Efficient time sequence analysis method and device based on lightweight convolutional neural network, and medium

Abstract

The invention discloses an efficient time sequence analysis method and device based on a convolutional neural network and a medium. Time sequence modeling analysis results for different downstream tasks are obtained through the steps of original data reading, sample standardization, adaptive time-frequency conversion, frequency domain component extraction and fitting, time domain data reconstruction and the like. In the time sequence analysis method, a time-frequency conversion and time domain data reconstruction step is realized through a dynamically adjustable wavelet analysis component, and a frequency domain data fitting step efficiently models distribution of frequency components in original data through a convolutional neural network with low time and space complexity. Raw data reading and sample standardization provide a starting point for time series analysis for different tasks. The method is based on data driving, has no special requirement for data, is high in universality, can maintain high modeling precision while greatly reducing time cost and hardware resource overhead, and has high theoretical property and practicability.

Description

Technical Field

[0001] The present invention relates to a time series analysis method, and particularly to a time series modeling and analysis method, device and medium based on a lightweight convolutional neural network. Background Art

[0002] Time series analysis is an important branch in the fields of statistics and data science. It mainly studies data points arranged in chronological order to identify trends, seasonality, periodicity, and random fluctuations in the data. With the advent of the big data era, time series analysis has been increasingly widely applied in fields such as finance, meteorology, healthcare, and economic forecasting. Existing time series analysis methods mostly adopt statistical models and deep learning models. Among them, the statistical models are ARIMA and VAR, and these models all regard time series as variables for solution based on statistical analysis. However, the data in real scenarios does not follow a linear distribution, that is, a linear solution cannot be obtained for this data. At the same time, solving a high-order non-linear equation system is a thorny problem. The representatives of deep learning models are the sequence models LSTM and Transformer. Such models are completely data-driven and achieve iterative optimization based on the loss function to complete the fitting of the original data, thus overcoming the defect that traditional methods cannot analyze non-linear data. However, both LSTM and Transformer need to maintain long-distance historical information to establish long context associations, which makes the time complexity and space complexity of such models increase quadratically with the increase in the length of the input data and are not applicable to large-scale data sets. By transferring the time-domain sequence to the frequency domain, the periodic components of the time series can be efficiently modeled in a method with low computational time complexity and low hardware overhead, and long-distance modeling can be achieved without maintaining long historical information. Summary of the Invention

[0003] The purpose of the present invention is to improve and standardize the deficiencies in existing research and technologies, and propose an efficient time series analysis method based on a lightweight convolutional neural network. This method provides a new research perspective for time series analysis. It can achieve higher modeling accuracy than existing technologies on the basis of significantly reducing the required time and space overhead compared with existing research, and has higher practical value. Moreover, as an efficient method, the time and space complexity of the proposed algorithm will not increase sharply when dealing with the continuous expansion of the input data scale, and the scalability of the method application is high.

[0004] The purpose of the present invention is achieved by the following technical solutions:

[0005] An efficient time series analysis method based on a lightweight convolutional neural network, comprising: Step 1: Build an efficient time series analysis model with an efficient convolutional neural network as the feature extractor and an adaptive wavelet analysis component as the time-frequency conversion component; Step 2: According to different types of time series downstream tasks, read the time series in the calibration unit of the corresponding time series dataset; Step 3: Use the time series dataset in Step 2 to eliminate the dimensions of different variables for each sequence variable to standardize the data; Step 4: Input the standardized data in Step 3 into the efficient time series analysis model constructed in Step 1, and perform adaptive wavelet decomposition on the standardized data through the wavelet analysis component to convert the time-domain data to the frequency domain to obtain the global frequency-domain distribution; Step 5: Separate the global frequency-domain distribution into high-frequency components and low-frequency components, and use the efficient convolutional neural network to fit the component data distribution and generate high-frequency representations and low-frequency representations; Step 6: Input the high-frequency representations and low-frequency representations into the wavelet analysis component in Step 4 to achieve adaptive wavelet reconstruction and restore them to time-domain representations; Step 7: Map the restored time-domain representations in Step 6 to an output matrix by the output layer, calculate the error with the true value through the mean square error loss function, then calculate the derivative gradient of the error and perform backpropagation to optimize both the efficient convolutional neural network and the wavelet analysis component simultaneously; Step 8: According to different types of time series downstream tasks, read the corresponding time series dataset into the efficient time series analysis model, perform dimensionality increase and representation operations on the read features, and then output the modeling results, which are mapped to the final analysis results by the output layer.

[0006] Further, Step 2 includes: Record the number of single-time data records at a certain moment. The record entries at this moment include timestamps and sequence variables, and combine multiple data records at different moments into a time series dataset.

[0007] Further, Step 3 includes: Denote a certain time series variable, calculate its mathematical expectation and standard deviation, and after obtaining the mathematical expectation and standard deviation, divide the original sequence minus the mathematical expectation by the standard deviation to obtain the standardized sequence.

[0008] Further, Step 4 includes: The wavelet analysis component receives the original data as a single parameter for frequency-domain conversion, and the conversion method is: Initialize the high-pass decomposition filter function and the low-pass decomposition filter function; Record a time series variable. After shifting the high-pass decomposition filter function backward by the sequence length in terms of time units, multiply it with this time series variable. After shifting the low-pass decomposition filter function backward by the sequence length in terms of units, also multiply it with this time series variable. After completing one operation, reset the high-pass decomposition filter function and the low-pass decomposition filter function back to their original positions, then shift backward by one less than the variable length in terms of time units and repeat the above operations until all the product results of the high-pass decomposition filter and the low-pass decomposition filter are obtained. Then, accumulate the results respectively to obtain the high-frequency component spectrum and the low-frequency component spectrum.

[0009] Further, step 5 includes: Step 5.1, after obtaining the high-frequency component spectrum and the low-frequency component spectrum, perform characterization modeling through two independent efficient convolutional neural networks respectively. The operation process of convolution is as follows: Randomly initialize the high-frequency component convolution kernel and the low-frequency component convolution kernel; From the high-frequency component spectrum and the low-frequency component spectrum obtained in step 4, shift the high-frequency component convolution kernel backward by the spectrum length in terms of time units and multiply it with the high-frequency component spectrum, shift the low-frequency component convolution kernel backward by the variable length in terms of units and multiply it with the low-frequency component spectrum. After completing one operation, reset the high-pass decomposition filter function and the low-pass decomposition filter function back to their original positions, then shift backward by one less than the spectrum length in terms of time units and repeat the above operations until all the product results of the low-frequency component convolution kernel and the high-frequency component convolution kernel are obtained. Then, accumulate the results respectively to obtain the primary non-linear high-frequency characterization and the primary non-linear low-frequency characterization; Step 5.2, after obtaining the preliminary characterization, enhance the non-linear expression ability of the characterization pair through the GELU activation function to obtain the final characterization pair. The calculation process is as follows: Integrate the exponential function from negative infinity to the characterization length and multiply it with the primary non-linear high-frequency characterization to obtain the final non-linear high-frequency characterization, integrate the exponential function from negative infinity to the characterization length and multiply it with the primary non-linear low-frequency characterization to obtain the final non-linear low-frequency characterization.

[0010] Further, step 6 includes: The wavelet analysis component receives the non-linear high-frequency characterization and the non-linear low-frequency characterization obtained in step 5 as parameters to reconstruct the characterization information in the time domain. The conversion method is as follows: Initialize the high-pass reconstruction filter function and the low-pass reconstruction filter function; Denote the restored time-domain representation, and its calculation process is as follows: Shift the high-pass reconstruction filter function backward by the representation length units and multiply it by the high-frequency representation, shift the low-pass reconstruction filter function backward by the representation length units and multiply it by the low-frequency representation. After one operation, reset the high-pass reconstruction filter and the low-pass reconstruction filter back to their original positions, then shift backward by one unit less than the representation length and repeat the above operation until all the product results of the high-pass reconstruction filter and the low-pass reconstruction filter are obtained. After that, sum up all the results and multiply by 2 to get the restored time-domain representation.

[0011] Further, step 7 includes: After obtaining the time-domain representation restored in step 6, calculate its mean square error with the true value, and update all the parameters in the wavelet analysis component and the efficient convolutional neural network according to the result of the mean square error. The calculation process of the mean square error is: For the predicted value of the time-domain representation at each time point, calculate the difference between it and the true value at that point. After obtaining all the differences at all time points, sum them up and divide the sum result by the total number of time points to get the final result.

[0012] The present invention also provides an efficient time series analysis device based on a lightweight convolutional neural network, including one or more processors for implementing an efficient time series analysis method based on a lightweight convolutional neural network as described above.

[0013] The present invention also provides a readable storage medium with a program stored thereon. When the program is executed by a processor, it implements an efficient time series analysis method based on a lightweight convolutional neural network as described above.

[0014] Compared with the prior art, the present invention has the following innovative advantages and remarkable effects:

[0015] 1) Using a brand-new frequency-domain time series modeling paradigm, it can quickly and efficiently analyze the periodic composition of a given time series, and map the modeling results in the frequency domain to the time domain to correspond to different types of downstream analysis tasks;

[0016] 2) For different types of time series data, the specific implementation methods of decomposition and reconstruction during time-frequency transformation can vary, having strong flexibility and scalability;

[0017] 3) While improving the modeling accuracy of various time series tasks, it greatly reduces the required time and hardware overhead, and has good practicality and deployability in real-world scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flowchart of a time series modeling and analysis method based on a lightweight convolutional neural network of the present invention.

[0019] Figure 2It is the amplitude map of the high-frequency component obtained after the adaptive time-frequency conversion in step 4 of the embodiment to which a time series modeling and analysis method based on a lightweight convolutional neural network of the present invention is applied.

[0020] Figure 3 It is the amplitude map of the low-frequency component obtained after the adaptive time-frequency conversion in step 4 of the embodiment to which a time series modeling and analysis method based on a lightweight convolutional neural network of the present invention is applied.

[0021] Figure 4 It is the characterization map of the high-frequency component obtained by fitting according to the high-frequency component data distribution in step 5 of a time series modeling and analysis method based on a lightweight convolutional neural network of the present invention.

[0022] Figure 5 It is the characterization map of the low-frequency component obtained by fitting according to the low-frequency component data distribution in step 5 of a time series modeling and analysis method based on a lightweight convolutional neural network of the present invention.

[0023] Figure 6 It is the prediction curve obtained by reconstructing the obtained low-frequency characterization and high-frequency characterization into time-domain data in step 6 of a time series modeling and analysis method based on a lightweight convolutional neural network of the present invention.

[0024] Figure 7 It is the structural schematic diagram of an efficient time series analysis device based on a lightweight convolutional neural network of the present invention. Specific embodiments

[0025] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0026] Please refer to Figure 1 , an efficient time series analysis method based on a lightweight convolutional neural network, includes the following steps:

[0027] Step 1, construction of an efficient time series analysis model: According to the characteristics of time series data, select an efficient convolutional neural network as the feature extractor and an adaptive wavelet analysis component as the time-frequency conversion component to build an efficient time series analysis model, specifically including:

[0028] A convolutional neural network is a type of feedforward neural network that contains convolutional computations and has a deep structure, and is one of the representative algorithms of deep learning. Convolutional neural networks have the ability of feature learning and can perform translation-invariant classification on input information according to their hierarchical structure. Through the stacking of small-sized convolutional kernels, efficient convolutional modeling can be achieved.

[0029] Wavelet analysis is a method of localizing analysis in time and frequency. It gradually performs multi-scale refinement on a numerical sequence through the stretching and translation operations of basis functions, and finally achieves fine time division at high frequencies and fine frequency division at low frequencies. It can automatically adapt to the requirements of time-frequency signal analysis, and thus can focus on any details of the signal. In an efficient time series analysis model, by setting learnable basis functions, the basis functions can be adaptively optimized to the most suitable shape using the gradient descent algorithm.

[0030] Step 2, reading of original data: According to different types of downstream tasks of time series, read the time series in the calibration unit of the corresponding time series dataset, specifically including:

[0031] Denote the number of single-time data records at the th moment as pieces. The record entry at this moment contains the timestamp , the sequence variable . Denote the time series dataset as , where .

[0032] Step 3, sample standardization: Using the time series dataset in Step 2, calculate the mathematical expectation and the standard deviation for each sequence variable . Then, through and , scale to eliminate the dimensions of different variables in the original time series, so as to ensure data comparability.

[0033] Among them, denote a single time series variable as , calculate its mathematical expectation and the standard deviation . The calculation formulas are as follows:

[0034]

[0035] where k is the total number of samples and t is the time;

[0036] Standardize the sequence variable . The calculation formula is as follows:

[0037] .

[0038] In the formula, represents the standardized sequence.

[0039] Step 4, Adaptive time-frequency conversion: Input the data standardized in Step 3 into the efficient time series analysis model constructed in Step 1, and perform adaptive wavelet decomposition on the standardized data through the wavelet analysis component to convert the time-domain data into the frequency domain to obtain the global frequency-domain distribution , specifically including:

[0040] In the stage of adaptive wavelet decomposition, the wavelet analysis component receives the original data as a single parameter for frequency-domain conversion, and the conversion method is:

[0041] Initialize the high-pass decomposition filter function and the low-pass decomposition filter function ;

[0042] Denote the single time series variable as , and calculate its high-frequency component spectrum and low-frequency component spectrum , and their calculation formulas are as follows:

[0043]

[0044] In the formula, represents the discrete convolution operator, represents the translation step of the filter function.

[0045] Step 5, Frequency-domain distribution pattern fitting: First, perform a separation operation on the global frequency-domain distribution to obtain the high-frequency component and low-frequency component , and use the efficient convolutional neural network to fit the data distribution of and enhance the non-linear expression ability through the activation function, so as to obtain the spectrum representation pair after modeling , where represents the high-frequency representation, represents the low-frequency representation. Specifically including:

[0046] Step 5.1, In the stage of distribution pattern fitting, after obtaining the spectrum pair , perform representation modeling through the efficient convolutional neural network, and the operation process of convolution is:

[0047] Randomly initialize the high-frequency component convolution kernel and the low-frequency component convolution kernel ;

[0048] The high-frequency component spectrum and low-frequency component spectrum obtained from Step 4 , and its characterization calculation formula is as follows:

[0049]

[0050] In the formula, represents the primary non-linear high-frequency characterization, represents the primary non-linear low-frequency characterization.

[0051] Step 5.2, after obtaining the primary non-linear characterization pair, enhance the non-linear expression ability of the characterization pair through the GELU activation function to obtain the final characterization pair , and its calculation process is as follows:

[0052]

[0053]

[0054] In the formula, represents the GELU activation function.

[0055] Step 6, adaptive time-domain feature reconstruction: Use the frequency component characterization pair modeled in Step 5 as the initial value, and pass it as a parameter into the wavelet analysis component in Step 4 to achieve adaptive wavelet reconstruction and restore it to the time-domain characterization, specifically including:

[0056] In the stage of time-domain feature reconstruction, the wavelet analysis component receives the two spectral characterization pairs obtained in Step 5 as parameters to reconstruct the characterization information in the time domain, and the conversion method is:

[0057] Initialize the high-pass reconstruction filter function and the low-pass reconstruction filter function ;

[0058] Denote the restored time-domain characterization as , and its calculation formula is as follows:

[0059] .

[0060] Step 7, error backpropagation and optimization: After obtaining the restored time-domain characterization through Step 6, map the time-domain characterization into different shapes through the output layer according to different downstream analysis tasks, compare it with the true value, calculate the error through the mean square error loss function, and then backpropagate the error through the derivative gradient to optimize both the efficient convolutional neural network and the wavelet analysis component at the same time. Specifically including:

[0061] After obtaining the characterization result , calculate its mean square error with the true value , and update all the parameters in the wavelet analysis component and the convolutional network according to the result of the mean square error The calculation method of the mean square error is as follows:

[0062]

[0063] where k is the total number of samples, is the true value of the i-th sample, is the predicted value of the i-th sample.

[0064] Step 8, time series analysis: According to different types of time series downstream tasks, read the corresponding time series dataset into the trained efficient time series analysis model obtained in Step 7. After the efficient time series analysis model performs dimensionality increase and characterization operations on the read features, it outputs a modeling result, which is mapped by the output layer into the final analysis result.

[0065] See Figure 7 , an efficient time series analysis device based on a lightweight convolutional neural network provided by an embodiment of the present invention includes one or more processors for implementing an efficient time series analysis method based on a convolutional neural network in the above embodiment.

[0066] An embodiment of an efficient time series analysis device based on a lightweight convolutional neural network of the present invention can be applied to any device with data processing capabilities, and the any device with data processing capabilities can be a device or apparatus such as a computer. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a logically meaningful device, it is formed by the processor of any device with data processing capabilities where it is located reading the corresponding computer program instructions in the non-volatile memory into the memory for operation. From the hardware level, as Figure 7 shown, it is a hardware structure diagram of any device with data processing capabilities where an efficient time series analysis device based on a lightweight convolutional neural network of the present invention is located. In addition to Figure 7 the processor, memory, network interface, and non-volatile memory shown, generally according to the actual functions of the device where the embodiment is located, any device with data processing capabilities may also include other hardware, which will not be elaborated here.

[0067] The implementation processes of the functions and roles of each unit in the above device are specifically detailed in the implementation processes of the corresponding steps in the above method, which will not be elaborated here.

[0068] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, 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, it should be considered as the scope described in this specification.

[0069] An embodiment of the present invention also provides a readable storage medium, on which a program is stored. When the program is executed by a processor, it implements an efficient time series analysis method based on a convolutional neural network in the above embodiment.

[0070] The readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the foregoing embodiments, such as a hard disk or memory. The readable storage medium may also be an external storage device, such as a plug-in hard disk, a Smart Media Card (SMC), an SD card, a Flash Card, etc. equipped on the device. Further, the readable storage medium may also include both an internal storage unit of any device with data processing capabilities and an external storage device. The readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and can also be used to temporarily store data that has been output or will be output.

[0071] Embodiment

[0072] This embodiment processes the time series of the transformer oil temperature at a certain substation collected from a certain place from July 1, 2016 to June 26, 2018. The specific variables and relevant data information included in the data set are shown in Table 1:

[0073] Table 1 Partial time series data of transformer oil temperature at a certain place

[0074]

[0075] In this embodiment, the implementation data set of the multivariate time series analysis method is the above-mentioned time series data of the transformer oil temperature at a certain place, and the result of the method is the future prediction curve of the oil temperature variable. The detailed implementation steps are as follows:

[0076] Step 1, construction of an efficient time series analysis model: According to the characteristics of the time series data, select an efficient convolutional neural network as the feature extractor and an adaptive wavelet analysis component as the time-frequency conversion component to build an efficient time series analysis model. The adaptive wavelet analysis component is responsible for converting the time series into a frequency spectrum, and the efficient convolutional neural network characterizes the distribution pattern of the data in the frequency spectrum.

[0077] Step 2, according to the requirements obtained from the infectious disease trend curve, read the historical transformer tank temperature data in hours, the The number of single-hour data records at a moment is denoted as pieces. This information includes the recording timestamp , high useful load variable {v1}, high useless load variable {v2}, medium useless load variable {v3}, low useful load variable {v4}, low useless load variable {v5}, and oil temperature variable {v6}. Denote the time series data set as , where ; According to the data information provided in Table 1, it includes all the necessary information required in this step.

[0078] Step 3, utilize the time series data set in Step 2 , first calculate the mathematical expectation and standard deviation for each time series, and calculate the value based on this, and obtain the standardized result for each record. The results are shown in Table 2:

[0079] Table 2 Time series data after implementation of standardization (unitless after standardization)

[0080]

[0081] Step 4, input the data standardized in Step 3 into the efficient time series analysis model constructed in Step 1, perform adaptive wavelet decomposition on the standardized data through the wavelet analysis component to convert the time-domain data to the frequency domain to obtain the global frequency domain distribution, denoted as ; The calculation method is as follows:

[0082]

[0083] The high-frequency spectrum and low-frequency spectrum calculated from the original data in this embodiment are as Figure 2 and Figure 3 shown.

[0084] In the frequency domain distribution pattern fitting step, the specific fitting method is as follows:

[0085] a) Preset two independent convolutional neural networks and and randomly initialize the weights of the convolutional kernels of the neural network.

[0086] b) According to the frequency components calculated in Step 4 for , take them as parameters and pass them into the corresponding feature extraction networks respectively for data fitting and modeling. The convolutional neural network used in this embodiment is a one-dimensional convolution. The convolutional layer performs a convolution operation on the input data by sliding a fixed-size window, extracts the features within the window, and then maps these features to the next layer after enhancing their non-linear expression. It consists of multiple convolutional layers and pooling layers alternating, and finally uses a fully connected layer to map the extracted features to the output. During the training process, the neural network uses the backpropagation algorithm to update the model parameters to minimize the loss function. This step is carried out in two stages. The initial feature extraction calculation process in the first stage is as follows:

[0087]

[0088] Obtain the primary non-linear characterization pair After that, enhance its non-linear expression ability through the GELU operator, and its calculation method is as follows:

[0089]

[0090]

[0091] After the entire characterization modeling stage is completed, the fitted frequency component characterization pair with a fixed prediction duration T of 96 can be obtained , respectively as Figure 4 , Figure 5 shown.

[0092] Step 6, in the adaptive time-domain feature reconstruction step, use the obtained in step 5, and pass it as a parameter into the wavelet analysis component at the same time to reconstruct the frequency-domain periodic information into a time-domain signal ; The calculation method is as follows:

[0093]

[0094] After the reconstruction is completed, use to output the fitted features as a prediction curve for the evaluation and analysis of different downstream tasks. The results are as Figure 6 shown.

[0095] An efficient time series analysis method based on a lightweight convolutional neural network according to the present invention mainly obtains the final time series prediction curve through steps such as sample standardization, adaptive time-frequency conversion, frequency-domain distribution pattern fitting, and adaptive time-domain feature reconstruction. Figure 1 For the specific process of the time series efficient modeling method. The entire embodiment follows the Figure 1 shown process to process the transformer oil temperature dataset and finally obtain the temperature prediction curve. Figure 2 , Figure 3To calculate the high-frequency spectrum and low-frequency spectrum of the original data using the present invention, Figure 4 , Figure 5 To obtain the characterization by fitting the distribution pattern of the frequency information in the spectrum using the present invention, Figure 6 Then the final temperature prediction curve is obtained. By performing time-frequency transformation and in-frequency domain fitting on the oil temperature data, a prediction curve that is more in line with the actual situation can be obtained compared with the traditional time series analysis model. While meeting a higher degree of data authenticity, it also has the characteristics of high real-time performance, small resource overhead, and strong scalability, providing a more practical solution for subsequent related research on time series modeling tasks with high real-time requirements and other aspects.

[0096] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An efficient time series analysis method based on a lightweight convolutional neural network, characterized in that, Including: Step 1: Build an efficient time series analysis model with an efficient convolutional neural network as the feature extractor and an adaptive wavelet analysis component as the time-frequency conversion component. Step 2: According to different types of time series downstream tasks, read the time series in the calibration unit of the corresponding time series dataset. Step 3: Use the time series dataset in Step 2 to eliminate the dimensions of different variables for each sequence variable, standardizing the data. Step 4: Input the standardized data in Step 3 into the efficient time series analysis model constructed in Step 1, perform adaptive wavelet decomposition on the standardized data through the wavelet analysis component to convert the time-domain data to the frequency domain, and obtain the global frequency-domain distribution. Step 5: Separate the global frequency-domain distribution into high-frequency components and low-frequency components, and use the efficient convolutional neural network to fit the component data distribution and generate high-frequency representations and low-frequency representations. Step 6: Input the high-frequency representations and low-frequency representations into the wavelet analysis component in Step 4 to achieve adaptive wavelet reconstruction and restore them to time-domain representations. Step 7: Map the time-domain representations restored in Step 6 to an output matrix by the output layer, calculate the error together with the true value through the mean square error loss function, then calculate the derivative gradient of the error and perform backpropagation to optimize both the efficient convolutional neural network and the wavelet analysis component simultaneously. Step 8: According to different types of time series downstream tasks, read the corresponding time series dataset into the efficient time series analysis model, perform dimensionality increase and representation operations on the read features, and then output the modeling result, which is mapped to the final analysis result by the output layer.

2. The efficient time series analysis method based on a lightweight convolutional neural network according to claim 1, characterized in that Step 2 includes: Record the number of single-moment data records at a certain moment. The record entries at this moment include timestamps and sequence variables, and combine multiple data records at different moments into a time series dataset.

3. An efficient time series analysis method based on a lightweight convolutional neural network according to claim 1, characterized in that, Step 3 includes: Denote a certain time series variable, calculate its mathematical expectation and standard deviation, and after obtaining the mathematical expectation and standard deviation, divide the original sequence minus the mathematical expectation by the standard deviation to get the standardized sequence.

4. An efficient time series analysis method based on a lightweight convolutional neural network according to claim 1, characterized in that, Step 4 includes: The wavelet analysis component receives the original data as a single parameter for frequency-domain conversion. The conversion method is as follows: Initialize the high-pass decomposition filter function and the low-pass decomposition filter function. Denote a certain time series variable, multiply the high-pass decomposition filter function after shifting it backward by the sequence length in time units by this time series variable, and also multiply the low-pass decomposition filter function after shifting it backward by the sequence length in units by this time series variable. After completing one operation, reset the high-pass decomposition filter function and the low-pass decomposition filter function back to their original positions, then shift them backward by one less than the variable length in time units and repeat the above operation until all the product results of the high-pass decomposition filter and the low-pass decomposition filter are obtained. Then, sum the results respectively to get the high-frequency component spectrum and the low-frequency component spectrum.

5. An efficient time series analysis method based on a lightweight convolutional neural network according to claim 4, characterized in that Step 5 includes: Step 5.1: After obtaining the high-frequency component spectrum and the low-frequency component spectrum, perform representation modeling through two independent efficient convolutional neural networks respectively. The convolution operation process is as follows: Randomly initialize the high-frequency component convolution kernel and the low-frequency component convolution kernel. For the high-frequency component spectrum and the low-frequency component spectrum obtained in step 4, after shifting the high-frequency component convolution kernel backward by the length of the spectrum in terms of time units and multiplying it with the high-frequency component spectrum, and shifting the low-frequency component convolution kernel backward by the length of the variable in terms of units and multiplying it with the low-frequency component spectrum. After completing one operation, reset the high-pass decomposition filter function and the low-pass decomposition filter function back to their original positions, then shift backward by one less than the length of the spectrum in terms of time units and repeat the above operation. Until all the product results of the low-frequency component convolution kernel and the high-frequency component convolution kernel are obtained, then sum up the results respectively to get the primary non-linear high-frequency representation and the primary non-linear low-frequency representation; In step 5.2, after obtaining the preliminary representation, enhance the non-linear expression ability of the representation pair through the GELU activation function to obtain the final representation pair. The calculation process is as follows: Integrate the exponential function from negative infinity to the length of the representation and multiply it with the primary non-linear high-frequency representation to obtain the final non-linear high-frequency representation, and integrate the exponential function from negative infinity to the length of the representation and multiply it with the primary non-linear low-frequency representation to obtain the final non-linear low-frequency representation.

6. An efficient time series analysis method based on a lightweight convolutional neural network according to claim 5, characterized in that Step 6 includes: The wavelet analysis component receives the non-linear high-frequency representation and the non-linear low-frequency representation obtained in step 5 as parameters to reconstruct the representation information in the time domain. The conversion method is: Initialize the high-pass reconstruction filter function and the low-pass reconstruction filter function; Denote the restored time-domain representation. The calculation process is as follows: Shift the high-pass reconstruction filter function backward by the length of the representation in terms of units and multiply it with the high-frequency representation, shift the low-pass reconstruction filter function backward by the length of the representation in terms of units and multiply it with the low-frequency representation. After completing one operation, reset the high-pass reconstruction filter and the low-pass reconstruction filter back to their original positions, then shift backward by one less than the length of the representation in terms of units and repeat the above operation. Until all the product results of the high-pass reconstruction filter and the low-pass reconstruction filter are obtained, then sum up all the results and multiply by 2 to get the restored time-domain representation.

7. An efficient time series analysis method based on a lightweight convolutional neural network according to claim 1, characterized in that, Step 7 includes: After obtaining the time-domain representation restored in step 6, calculate the mean square error between it and the true value, and update all the parameters in the wavelet analysis component and the efficient convolutional neural network according to the result of the mean square error. The calculation process of the mean square error is as follows: For the predicted value of the time-domain representation at each time point, calculate the difference between it and the true value at that point, sum up the differences at all time points, and divide the sum by the total number of time points to get the final result.

8. An efficient time series analysis device based on a lightweight convolutional neural network, characterized in that, It includes one or more processors for implementing an efficient time series analysis method according to any one of claims 1-7 based on a lightweight convolutional neural network.

9. A readable storage medium, characterized in that, There is a program stored thereon, which when executed by the processor, implements an efficient time series analysis method according to any one of claims 1-7 based on a lightweight convolutional neural network.

Images