Meteorological prediction method and device based on double-channel modeling, equipment and storage medium

Abstract

The application relates to the technical field of weather prediction, in particular to a weather prediction method and device based on double-channel modeling, equipment and a storage medium. The method comprises the following steps: collecting historical weather observation data of a target area, the historical weather observation data being weather time series, and the weather comprising at least one of temperature, precipitation, humidity, wind speed and air pressure; performing feature enhancement on the weather time series; inputting the weather time series after the feature enhancement into a first feature modeling channel and a second feature modeling channel respectively, the first feature modeling channel being a multi-scale gated convolutional neural network, and the second feature modeling channel being an extended long short-term memory network; and performing fusion processing on output features of the first feature modeling channel and output features of the second feature modeling channel to generate weather data in a target time period. The application can capture long-period evolution rules of weather and take into account multi-scale and multi-variable weather features, so that the accuracy of weather prediction is greatly improved.

Description

Technical Field

[0001] This application relates to the technical field of weather forecasting, specifically to a weather forecasting method, apparatus, equipment, and storage medium based on dual-channel modeling. Background Technology

[0002] Time series forecasting is a crucial technique for modeling historical observation data and predicting future trends, with wide-ranging applications in weather forecasting, smart energy, and urban transportation. In weather forecasting scenarios, meteorological data not only contains multi-scale dynamic characteristics such as daily, weekly, and seasonal cycles, but also exhibits multi-variable coupled relationships such as temperature and air pressure, humidity and precipitation. Therefore, meteorological systems are typically complex nonlinear systems. Traditional statistical models (such as autoregressive moving averages and exponential smoothing) have limitations in handling nonlinear, non-stationary, and long-term dependent relationships, making it difficult to capture the complex dynamic characteristics of meteorological data.

[0003] In recent years, deep learning-based time series forecasting methods have made some progress in prediction accuracy by modeling complex temporal features using neural networks. However, in meteorological forecasting scenarios, there are still problems in taking into account the long-term evolution patterns of meteorological elements as well as multi-scale and multi-variable characteristics, resulting in lower accuracy in meteorological forecasts. Summary of the Invention

[0004] The purpose of this application is to overcome the shortcomings of existing meteorological forecasting scenarios, which still have the problem of difficulty in taking into account the long-term evolution law of meteorological elements and the multi-scale and multi-variable characteristics, resulting in low accuracy of meteorological forecasts.

[0005] Therefore, this application provides a meteorological forecasting method based on dual-channel modeling, including the following steps: Historical meteorological observation data of the target area are collected. The historical meteorological observation data is a meteorological time series, and the meteorological data includes at least one of temperature, precipitation, humidity, wind speed and air pressure. Feature enhancement for meteorological time series; The enhanced meteorological time series is input into the first feature modeling channel and the second feature modeling channel, respectively. The first feature modeling channel is a multi-scale gated convolutional neural network, and the second feature modeling channel is an extended long short-term memory network. The output features of the first feature modeling channel and the output features of the second feature modeling channel are fused together to generate meteorological data for the target time period.

[0006] Furthermore, the steps for processing meteorological time series in the first feature modeling channel include: Up-dimensionalize the meteorological time series with enhanced features; Multi-scale feature extraction of the upgraded meteorological time series based on multi-scale convolution; Learnable gated weights are generated for each convolutional channel to achieve adaptive weighted fusion of multi-scale features; The fused multi-scale features are post-processed, including normalization and random deactivation. The dimensions of the multi-scale features after recovery and post-processing.

[0007] Furthermore, the multi-scale convolution in the step of extracting multi-scale features from the upgraded meteorological time series based on multi-scale convolution is a 2D convolution. The 2D convolution uses a large kernel decomposition strategy to decompose a single large convolution into two small cascaded convolutions.

[0008] Furthermore, the meteorological forecasting method based on dual-channel modeling also includes a step of preprocessing meteorological time series, wherein the preprocessing includes outlier detection processing, missing value repair processing, and normalization processing.

[0009] Furthermore, the missing value repair process employs a neighboring observation interpolation method with learnable weights, based on formula 1, which is: Formula 1, In Equation 1 , The time coordinates of the two most recent known data points before and after the missing value. , They are respectively with , Meteorological element values ​​at the corresponding time points, For learnable weights, The time coordinates of the missing points for Interpolation results of meteorological element values ​​at the time coordinate. < < .

[0010] Furthermore, the feature enhancement of the meteorological time series includes: The time-domain signal of the meteorological time series is converted into a frequency-domain signal by Fourier transform, and learnable weights are generated for the frequency-domain signal. Calculate the energy at each frequency point and normalize the energy at each frequency point; Frequency domain signals are filtered based on energy thresholds and energy values ​​at each frequency point; The filtered frequency domain signal is enhanced, and the enhanced meteorological time series is obtained by mapping the frequency domain signal back to the time domain signal based on the inverse Fourier transform.

[0011] Furthermore, the extended long short-term memory network includes an input linear layer, a single-layer extended long short-term memory network unit, and an output linear layer.

[0012] This application also provides a weather forecasting device based on dual-channel modeling, comprising: A meteorological data acquisition module is used to collect historical meteorological observation data of a target area. The historical meteorological observation data is a meteorological time series, and the meteorological data includes at least one of temperature, precipitation, humidity, wind speed, and air pressure. The feature enhancement module is used to enhance the features of meteorological time series. The feature input module is used to input the feature-enhanced meteorological time series into the first feature modeling channel and the second feature modeling channel, respectively. The first feature modeling channel is a multi-scale gated convolutional neural network, and the second feature modeling channel is an extended long short-term memory network. The feature fusion module is used to fuse the output features of the first feature modeling channel and the output features of the second feature modeling channel to generate meteorological data for the target time period.

[0013] This application also provides a weather forecasting device based on dual-channel modeling, including a processor and a memory, wherein the memory stores a computer program and the processor executes the aforementioned weather forecasting method based on dual-channel modeling.

[0014] This application also provides a storage medium storing a program, which, when executed by a processor, is used to implement the aforementioned meteorological forecasting method based on dual-channel modeling.

[0015] Furthermore, The technical solution of this application has at least the following advantages: 1. The meteorological forecasting method based on dual-channel modeling in this application can capture the long-term evolution of meteorological patterns and take into account multi-scale and multi-variable meteorological characteristics through the synergistic effect of the first feature modeling channel and the second feature modeling channel, thereby greatly improving the accuracy of meteorological forecasting.

[0016] 2. The multi-scale gated convolutional neural network of this application improves prediction accuracy while controlling computational complexity through a large kernel decomposition strategy, greatly reducing the computational load in weather forecasting. In addition, the multi-scale gated convolutional neural network can comprehensively understand the multi-scale correlation between meteorological elements through convolutional kernels with different receptive fields and adaptive gating mechanisms, which is particularly effective in improving the prediction accuracy of highly coupled meteorological elements (such as humidity and precipitation).

[0017] 3. This application improves the accuracy of weather forecasting by accurately masking low-energy noise components and preserving key meteorological features through feature enhancement of meteorological time series. Furthermore, the use of neighboring observation point interpolation with learnable weights to repair missing values ​​more reasonably restores the dynamic trend of missing segments, further enhancing the accuracy of weather forecasting. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart of the meteorological forecasting method based on dual-channel modeling in Embodiment 1 of this application; Figure 2 This is a flowchart of the meteorological time series processing based on the first feature modeling channel in Embodiment 1 of this application; Figure 3 This is a structural block diagram of the meteorological forecasting device based on dual-channel modeling in Embodiment 2 of this application; Figure 4 This is a structural block diagram of the meteorological forecasting device based on dual-channel modeling in Embodiment 3 of this application. Detailed Implementation

[0020] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] In the description of this application, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0022] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal connection of two components; and they can refer to a wireless connection or a wired connection. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0023] Furthermore, the technical features involved in the different embodiments of this application described below can be combined with each other as long as they do not conflict with each other.

[0024] Example 1

[0025] Example 1 provides a weather forecasting method based on dual-channel modeling. Figure 1 This is a flowchart of the meteorological forecasting method based on dual-channel modeling in Embodiment 1 of this application. (Refer to...) Figure 1 The meteorological forecasting method based on dual-channel modeling includes the following steps: S1. Collect historical meteorological observation data of the target area. The historical meteorological observation data is a meteorological time series, and the meteorological data includes at least one of temperature, precipitation, humidity, wind speed and air pressure. Specifically, the time granularity in meteorological time series can be at the minute, hour, or day level.

[0026] Furthermore, after step S1, a step of preprocessing the meteorological time series is also included, wherein the preprocessing includes outlier detection processing, missing value repair processing, and normalization processing.

[0027] Specifically, outlier detection and processing include: using the interquartile range (IQR) method to identify outliers in meteorological time series, that is, calculating the first quartile Q1 and the third quartile Q3 for any meteorological series, and setting the outlier threshold as [Q1-1.5×IQR, Q3+1.5×IQR], where IQR is the difference between Q3 and Q1; those exceeding the outlier threshold range are outliers, and the moving average method is used to replace outliers.

[0028] Specifically, missing value repair is performed on the meteorological time series after outlier detection. The missing value repair method adopts the neighboring observation interpolation method with learnable weights, based on Equation 1, which is: Formula 1, In Equation 1 , The time coordinates of the two most recent known data points before and after the missing value. , They are respectively with , Meteorological element values ​​at the corresponding time points, For learnable weights, The initial value is set to 1.0. The time coordinates of the missing points for Interpolation results of meteorological element values ​​at the time coordinate. < < .

[0029] Specifically, the restored meteorological time series is subjected to Z-score normalization to obtain a standardized meteorological time series.

[0030] Furthermore, preprocessing includes constructing sample windows, where each sample group consists of continuous historical meteorological time series data, used to predict meteorological changes over several future time steps. Specifically, a sliding window method is used to segment the standardized meteorological time series to ensure the continuity and integrity of the data in the time dimension. The historical window length is set to... (For example, 720 hours corresponds to 30 days), the prediction window length is (e.g., 24 hours, 48 ​​hours, or 72 hours), meaning each sample meteorological time series contains... Meteorological data for a continuous historical time step, tagged as subsequent. Target meteorological data for each time step.

[0031] S2. Enhance the features of meteorological time series; The feature enhancement of meteorological time series includes: S21. Convert the time-domain signal of the meteorological time series into a frequency-domain signal by performing a Fourier transform, and generate learnable weights for the frequency-domain signal; specifically, convert the preprocessed sample meteorological time series into a frequency-domain signal by performing a Fourier transform.

[0032] S22. Calculate the energy at each frequency point and normalize the energy at each frequency point; Specifically, the formula for normalizing the energy at each frequency point is Equation 2, which is: Equation 2, In Equation 2 The energy at the current frequency point, The energy is the normalized value at the current frequency. The median energy across all frequencies. This is a temperature scaling parameter used to control the smoothness of the normalized result. To prevent learnable parameters from having a denominator of 0, The initial value is set to 1e-6.

[0033] S23. Filter frequency domain signals based on energy threshold and energy value at each frequency point; Specifically, an energy threshold is set, which is adaptively learned through training data, with an initial value set to... The mean value is used to generate a Boolean matrix as a mask based on the energy threshold and the energy value at each frequency point. Remove frequency points with energy values ​​below the energy threshold and retain frequency points with energy values ​​greater than or equal to the energy threshold.

[0034] S24. Enhance the selected frequency domain signal and map the frequency domain signal back to the time domain signal based on the inverse Fourier transform to obtain the enhanced meteorological time series. Specifically, the formula used for enhancing the filtered frequency domain signal is Equation 3, which is: Formula 3, In Equation 3, For the enhanced frequency domain signal, The frequency domain signal after masking is... Calculated according to Equation 4, Based on Equation 5, Equation 4 is calculated as follows: Equation 4, In Equation 4, For channel matrix multiplication, The frequency domain signal after Fourier transform. For the generated mask, For learnable complex weights, Equation 5 is: Equation 5, In Equation 5, For channel matrix multiplication, The frequency domain signal after Fourier transform. For learnable complex weights; The enhanced frequency domain signal is mapped back to the time domain signal through inverse Fourier transform, thus obtaining the noise-suppressed meteorological time series, which is the feature-enhanced meteorological time series.

[0035] In some embodiments, the feature enhancement step is an adaptive enhancement step, which includes: The enhanced meteorological time series is then decomposed and pieced together; specifically, the steps for decomposing and pieced together the enhanced meteorological time series include: The sliding window averaging method is used to decompose the noise-enhanced meteorological time series into trend terms. and seasonal items Among them, the trend item It is all time steps ( ∈[0, -1]) The set, The calculation is based on Equation 6, which is: Formula 6, In Equation 6, Trend term in the The value at each time step, The value represents the sliding window size. , For time step index, For the feature-enhanced time series in the 1st The value at each time step, For the time step index of the summation process, The meteorological time series after feature enhancement; Seasonal items The calculation is based on Equation 7, which is: Formula 7; In Equation 7, The meteorological time series after feature enhancement; Specifically, the decomposed trend items and seasonal items The meteorological time series containing seasonal and trend terms is spliced ​​using a feature splicing function. The spliced ​​meteorological time series is then updated by repeating steps 21-24.

[0036] Feature enhancement of meteorological time series data can accurately mask low-energy noise components while preserving key meteorological features, thereby improving the accuracy of weather forecasts. Furthermore, using neighboring observation point interpolation with learnable weights to repair missing values ​​can more reasonably restore the dynamic trends of missing segments, further enhancing the accuracy of weather forecasts.

[0037] S3. Input the feature-enhanced meteorological time series into the first feature modeling channel and the second feature modeling channel respectively. The first feature modeling channel is a multi-scale gated convolutional neural network, and the second feature modeling channel is an extended long short-term memory network (xLSTM). Specifically, Figure 2 This is a flowchart of processing meteorological time series based on the first feature modeling channel in Embodiment 1 of this application, with reference to... Figure 2 The steps for processing meteorological time series in the first feature modeling channel include: S311. Up-dimensionalize the meteorological time series after feature enhancement; Specifically, the unsqueeze method is used to upscale the feature-enhanced meteorological time series to an adaptive 2D convolution.

[0038] S312. Multi-scale feature extraction of the upgraded meteorological time series based on multi-scale convolution; Specifically, the multi-scale convolution in step S312 is a 2D convolution, which uses a large kernel decomposition strategy to decompose a single large convolution into two small concatenated convolutions. Specifically, in multi-scale convolution, three convolution channels with different receptive fields are set, and the convolution kernel sizes can be 1×3, 1×5, and 1×7, respectively, so as to extract multiple periodic information in meteorological time series. Furthermore, the large kernel decomposition strategy, based on Equation 8, replaces convolutions with larger kernels with two smaller concatenated convolutions. Equation 8 is as follows: Formula 8, In Equation 8, For a convolution with a kernel of 1×K, For a convolution with a kernel of 1×m, This represents a convolution with a kernel of 1×n. .

[0039] S313. Generate learnable gated weights for each convolutional channel to achieve adaptive weighted fusion of multi-scale features; Specifically, learnable gating weights are assigned to each convolutional channel based on the sigmoid gating mechanism. The gating weights are calculated using Equations 9 and 10, where Equation 9 is: Formula 9, Equation 10 is: Formula 10, In Equations 9 and 10, For the first The weights of each convolutional channel, To input the meteorological time series after feature enhancement processing via multi-scale convolutional channels, It is the sigmoid activation function. The first concatenated convolution obtained by decomposing a certain large kernel convolution is the first convolution. indivual, The learnable gated weight matrix, For the first Learnable gating weights for each convolutional channel; Based on learnable gating weights, an element-wise weighting method is used to achieve weighted fusion of multi-scale features, thereby outputting fused features that take into account multi-scale information.

[0040] S314. Post-process the fused multi-scale features, including normalization and random deactivation. Specifically, to further improve the training stability and generalization ability of the model, the multi-scale features after fusion are post-processed. The model in this application consists of a first feature modeling channel, a second feature modeling channel, and a part that fuses the output features of the two channels. Dropout technology is used for random deactivation, with a dropout rate of 0.5.

[0041] S315, Dimensions of multi-scale features after recovery and post-processing; Specifically, the squeeze method is used to reduce the dimensionality of the post-processed multi-scale features back to their original state before dimensionality increase.

[0042] In some embodiments, after dimensionality reduction, a 1D convolution with a kernel size of 1 is used to shape the processed meteorological time series into the desired shape, which is then used as the output of a multi-scale gated convolutional neural network.

[0043] In some embodiments, the Extended Long Short-Term Memory (xLSTM) network consists of an input linear layer, a single-layer xLSTM unit, and an output linear layer sequentially. The input linear layer first projects the lookback window dimension of the input feature-enhanced meteorological time series to a lower-dimensional space (e.g., 256 dimensions) to compress temporal information and improve feature representation. Then, a linear mapping transforms the time dimension of the sequence to the desired prediction length (e.g., 720 dimensions), achieving a continuous mapping from historical information to future time steps.

[0044] The meteorological forecasting method based on dual-channel modeling described above can capture the long-term evolution of meteorological patterns and take into account multi-scale and multi-variable meteorological characteristics through the synergistic effect of the first and second feature modeling channels, thereby greatly improving the accuracy of meteorological forecasting.

[0045] Furthermore, multi-scale gated convolutional neural networks (MLNs) improve prediction accuracy while controlling computational complexity through a large kernel decomposition strategy, significantly reducing the computational load in weather forecasting. In addition, MNNs, through convolutional kernels with different receptive fields and adaptive gating mechanisms, can comprehensively understand the multi-scale relationships between meteorological elements, with particularly significant improvements in prediction accuracy for highly coupled meteorological elements (such as humidity and precipitation).

[0046] S4. The output features of the first feature modeling channel and the output features of the second feature modeling channel are fused together to generate meteorological data for the target time period.

[0047] Specifically, an adaptive weighted fusion method is used to fuse the output features of the first feature modeling channel and the output features of the second feature modeling channel, thereby outputting meteorological data within the target time period, which is the meteorological forecast result.

[0048] Example 2

[0049] Example 2 provides a weather forecasting device based on dual-channel modeling. Figure 3 This is a structural block diagram of the meteorological forecasting device based on dual-channel modeling in Embodiment 2 of this application. (Refer to...) Figure 3 The meteorological forecasting device based on dual-channel modeling includes a meteorological acquisition module 301, a feature enhancement module 303, a feature input module 304, and a feature fusion module 305. The meteorological acquisition module 301 collects historical meteorological observation data of the target area. The historical meteorological observation data is a meteorological time series, and the meteorological data includes at least one of temperature, precipitation, humidity, wind speed, and air pressure. The feature enhancement module 303 enhances the features of the meteorological time series. The feature input module 304 inputs the feature-enhanced meteorological time series into a first feature modeling channel and a second feature modeling channel, respectively. The first feature modeling channel is a multi-scale gated convolutional neural network, and the second feature modeling channel is an extended long short-term memory network. The feature fusion module 305 fuses the output features of the first feature modeling channel and the output features of the second feature modeling channel to generate meteorological data for the target time period. Furthermore, the meteorological forecasting device based on dual-channel modeling also includes a preprocessing module 302, which performs outlier detection, missing value repair, and normalization processing on the meteorological time series.

[0050] Example 3

[0051] Example 3 also provides a weather forecasting device based on dual-channel modeling. Figure 4 This is a structural block diagram of the meteorological forecasting device based on dual-channel modeling in Embodiment 3 of this application. (Refer to...) Figure 4 The weather forecasting device based on dual-channel modeling includes a processor 410 and a memory 420.

[0052] Processor 410 may include one or more processing cores. Processor 410 connects to various parts of the device using various interfaces and lines, and performs various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 420, and by calling data stored in memory 420. Optionally, processor 410 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). Processor 410 may integrate one or more of a Central Processing Unit (CPU) and a modem. The CPU primarily handles the operating system and applications; the modem is used for wireless communication. It is understood that the modem may also not be integrated into processor 410 and may be implemented as a separate chip.

[0053] Optionally, when the processor 410 executes the program instructions in the memory 420, it implements the weather forecasting method based on dual-channel modeling provided in the above-described method embodiments.

[0054] The memory 420 may include random access memory (RAM) or read-only memory. Optionally, the memory 420 may include a non-transitory computer-readable storage medium. The memory 420 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 420 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function, instructions for implementing the various method embodiments described above, etc.; the data storage area may store data created according to the use of the device, etc.

[0055] Optionally, this application also provides a storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the weather forecasting method based on dual-channel modeling provided in the above method embodiments.

[0056] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this application.

Claims

1. A meteorological forecasting method based on dual-channel modeling, characterized in that, Includes the following steps: Historical meteorological observation data of the target area are collected. The historical meteorological observation data is a meteorological time series, and the meteorological data includes at least one of temperature, precipitation, humidity, wind speed and air pressure. Feature enhancement for meteorological time series; The enhanced meteorological time series is input into the first feature modeling channel and the second feature modeling channel, respectively. The first feature modeling channel is a multi-scale gated convolutional neural network, and the second feature modeling channel is an extended long short-term memory network. The output features of the first feature modeling channel and the output features of the second feature modeling channel are fused together to generate meteorological data for the target time period. The steps for processing meteorological time series in the first feature modeling channel include at least: Up-dimensionalize the meteorological time series with enhanced features; Multi-scale feature extraction of the upgraded meteorological time series based on multi-scale convolution; Learnable gated weights are generated for each convolutional channel to achieve adaptive weighted fusion of multi-scale features; Recover the dimensionality of multi-scale features; The multi-scale convolution in the step of multi-scale feature extraction of the upgraded meteorological time series based on multi-scale convolution is a 2D convolution. The 2D convolution uses a large kernel decomposition strategy to decompose a single large convolution into two small cascaded convolutions. The extended long short-term memory network includes an input linear layer, a single-layer extended long short-term memory network unit, and an output linear layer. The feature enhancement of meteorological time series includes: The time-domain signal of the meteorological time series is converted into a frequency-domain signal by Fourier transform, and learnable weights are generated for the frequency-domain signal. Calculate the energy at each frequency point and normalize the energy at each frequency point; Frequency domain signals are filtered based on energy thresholds and energy values ​​at each frequency point; The filtered frequency domain signal is enhanced, and the frequency domain signal is mapped back to the time domain signal based on the inverse Fourier transform to obtain the enhanced meteorological time series; An adaptive weighted fusion method is used to fuse the output features of the first feature modeling channel and the output features of the second feature modeling channel, thereby outputting meteorological data within the target time period, which is the meteorological forecast result.

2. The meteorological forecasting method based on dual-channel modeling according to claim 1, characterized in that, The steps for processing meteorological time series in the first feature modeling channel also include: The fused multi-scale features are post-processed, including normalization and random deactivation. The dimension of the recovered multi-scale features is the dimension of the recovered multi-scale features after post-processing.

3. The meteorological forecasting method based on dual-channel modeling according to claim 1, characterized in that, It also includes a step of preprocessing meteorological time series, which includes outlier detection, missing value repair, and normalization.

4. The meteorological forecasting method based on dual-channel modeling according to claim 3, characterized in that, The missing value repair process employs a neighboring observation interpolation method with learnable weights, based on formula 1, which is: Formula 1, In Equation 1 , The time coordinates of the two most recent known data points before and after the missing value. , They are respectively with , Meteorological element values ​​at the corresponding time points, For learnable weights, The time coordinates of the missing points for Interpolation results of meteorological element values ​​at the time coordinate. < < .

5. A weather forecasting device based on dual-channel modeling, characterized in that, include: A meteorological data acquisition module is used to collect historical meteorological observation data of a target area. The historical meteorological observation data is a meteorological time series, and the meteorological data includes at least one of temperature, precipitation, humidity, wind speed, and air pressure. The feature enhancement module is used to enhance the features of meteorological time series. The feature input module is used to input the feature-enhanced meteorological time series into the first feature modeling channel and the second feature modeling channel, respectively. The first feature modeling channel is a multi-scale gated convolutional neural network, and the second feature modeling channel is an extended long short-term memory network. The feature fusion module is used to fuse the output features of the first feature modeling channel and the output features of the second feature modeling channel to generate meteorological data for the target time period. The steps for processing meteorological time series in the first feature modeling channel include at least: Up-dimensionalize the meteorological time series with enhanced features; Multi-scale feature extraction of the upgraded meteorological time series based on multi-scale convolution; Learnable gated weights are generated for each convolutional channel to achieve adaptive weighted fusion of multi-scale features; Recover the dimensionality of multi-scale features; The multi-scale convolution in the step of multi-scale feature extraction of the upgraded meteorological time series based on multi-scale convolution is a 2D convolution. The 2D convolution uses a large kernel decomposition strategy to decompose a single large convolution into two small concatenated convolutions. The extended long short-term memory network includes an input linear layer, a single-layer extended long short-term memory network unit, and an output linear layer. The feature enhancement of meteorological time series includes: The time-domain signal of the meteorological time series is converted into a frequency-domain signal by Fourier transform, and learnable weights are generated for the frequency-domain signal. Calculate the energy at each frequency point and normalize the energy at each frequency point; Frequency domain signals are filtered based on energy thresholds and energy values ​​at each frequency point; The filtered frequency domain signal is enhanced, and the frequency domain signal is mapped back to the time domain signal based on the inverse Fourier transform to obtain the enhanced meteorological time series; An adaptive weighted fusion method is used to fuse the output features of the first feature modeling channel and the output features of the second feature modeling channel, thereby outputting meteorological data within the target time period, which is the meteorological forecast result.

6. A meteorological forecasting device based on dual-channel modeling, characterized in that, It includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the weather forecasting method based on dual-channel modeling as described in any one of claims 1-4.

7. A storage medium, characterized in that, The storage medium stores a program, which, when executed by a processor, is used to implement the meteorological forecasting method based on dual-channel modeling as described in any one of claims 1-4.

Images