Multimodal time alignment method and device for power grid meteorological load joint prediction

Abstract

The application provides a power grid meteorological load combined prediction method and device based on multi-modal time sequence alignment. The method obtains power grid load data and power grid meteorological data, then performs time sequence alignment operation on the power grid load data and the power grid meteorological data at the same target time granularity, and generates multi-modal time sequence feature data according to the power grid load data and the power grid meteorological data aligned at the target time granularity, so as to train a prediction model for power grid load prediction by using the multi-modal time sequence feature data. In this way, the time sequence alignment operation can retain the original time characteristics of data with large fluctuations as much as possible while ensuring the time consistency of multi-modal time sequence data, thereby reducing the further false fluctuations and errors introduced by subsequent interpolation or completion, and improving the accuracy of the trained prediction model.

Description

Technical Field

[0001] This application relates to data processing technology, and in particular to a method and apparatus for joint forecasting of power grid meteorological loads using multimodal time-series alignment. Background Technology

[0002] With the continuous expansion of the power grid and the increase in the proportion of renewable energy integration, the operating environment of the power grid is becoming increasingly complex, placing higher demands on the accuracy of short-term and medium-to-long-term load forecasting. In order to improve forecast accuracy, recent research and applications have increasingly incorporated meteorological factors, such as temperature, humidity, wind speed, and precipitation, and combined them with power grid load data for joint modeling and forecasting.

[0003] In practice, power grid load data and meteorological data are often recorded at different sampling frequencies, leading to inconsistencies in their time scales. To address this issue, existing technologies typically employ a fixed-granularity unification method to align multimodal time series data.

[0004] However, when dealing with data with large fluctuations, the fixed granularity uniform method may introduce non-real fluctuation peaks or valleys due to resampling, resulting in noise amplification, destroying the original time-series fluctuation characteristics of the data, and thus introducing false fluctuations, which in turn affects the accuracy of the trained prediction model. Summary of the Invention

[0005] This application provides a method and apparatus for joint forecasting of power grid meteorological load using multimodal time-series alignment. While ensuring the temporal consistency of multimodal time-series data, the time-series alignment operation retains the original temporal characteristics of data with large fluctuations as much as possible, thereby reducing further spurious fluctuations and errors introduced by subsequent interpolation or completion, and thus improving the accuracy of the trained forecasting model.

[0006] Firstly, this application provides a multimodal time-series aligned joint forecasting method for power grid meteorological loads, including:

[0007] Acquire power grid load data and power grid meteorological data;

[0008] At the same target time granularity, a time-series alignment operation is performed on the power grid load data and the power grid meteorological data, wherein the time-series alignment operation is used to align the time granularity of the data with smaller data fluctuations in the power grid load data and the power grid meteorological data towards the time granularity of the data with larger data fluctuations.

[0009] Multimodal time-series feature data is generated based on the grid load data aligned at the target time granularity and the grid meteorological data, and the prediction model for grid load forecasting is trained using the multimodal time-series feature data.

[0010] Optionally, the time difference between the time granularity corresponding to the data with large fluctuations in the power grid load data and the target time granularity is smaller than the time difference between the time granularity corresponding to the data with small fluctuations and the target time granularity.

[0011] Optionally, the time-series alignment operation between the power grid load data and the power grid meteorological data at the same target time granularity includes:

[0012] The target time granularity is determined based on the power grid load data and the power grid meteorological data. The target time granularity is the time granularity corresponding to the data with large fluctuations in the power grid load data and the power grid meteorological data.

[0013] Only data with relatively small fluctuations in the power grid load data and the power grid meteorological data are time-aligned to the target time granularity.

[0014] Optionally, determining the target time granularity based on the power grid load data and the power grid meteorological data includes:

[0015] Based on the power grid load data, load fluctuation characteristic parameters for characterizing the degree of load fluctuation are determined, and based on the power grid meteorological data, meteorological fluctuation characteristic parameters for characterizing the degree of meteorological fluctuation are determined.

[0016] The target time granularity is determined by identifying the sampling rate corresponding to the larger fluctuation characteristic parameter among the load fluctuation characteristic parameters and the meteorological fluctuation characteristic parameters, and the data corresponding to the smaller fluctuation parameter is aligned according to the target time granularity.

[0017] Optionally, aligning the data corresponding to the other smaller fluctuation parameter according to the target time granularity includes:

[0018] If the original time granularity of the data corresponding to the smaller fluctuation parameter among the load fluctuation characteristic parameters and the meteorological fluctuation characteristic parameters is smaller than the target time granularity, then the corresponding data is downsampled according to the target time granularity.

[0019] If the original time granularity of the data corresponding to the smaller fluctuation parameter among the load fluctuation characteristic parameters and the meteorological fluctuation characteristic parameters is greater than the target time granularity, then the corresponding data is interpolated and completed according to the target time granularity.

[0020] Optionally, the load fluctuation characteristic parameter and / or the meteorological fluctuation characteristic parameter are determined based on the ratio between at least one of the data variance, root mean square error, or mean absolute difference and the corresponding preset indicator.

[0021] Optionally, acquiring power grid load data and power grid meteorological data includes:

[0022] The power grid load data at a first sampling frequency and the power grid meteorological data at a second sampling frequency are acquired, wherein the first sampling frequency and the second sampling frequency are different.

[0023] Secondly, this application provides a multimodal time-aligned power grid meteorological load joint forecasting device, characterized in that it includes:

[0024] The acquisition module is used to acquire power grid load data and power grid meteorological data;

[0025] The processing module is used to perform time-series alignment operations on the power grid load data and the power grid meteorological data at the same target time granularity. The time-series alignment operation is used to align the time granularity of the data with smaller data fluctuations in the power grid load data and the power grid meteorological data to the time granularity of the data with larger data fluctuations.

[0026] The training module is used to generate multimodal time-series feature data based on the aligned grid load data and grid meteorological data at the target time granularity, so as to train the prediction model for grid load forecasting using the multimodal time-series feature data.

[0027] The multimodal time-series aligned power grid and meteorological load joint forecasting method and apparatus provided in this application acquires power grid load data and power grid meteorological data, and then performs a time-series alignment operation on the power grid load data and power grid meteorological data at the same target time granularity. The time-series alignment operation aligns the time granularity of data with smaller fluctuations in the power grid load data and power grid meteorological data towards the time granularity of data with larger fluctuations. Then, multimodal time-series feature data is generated based on the aligned power grid load data and power grid meteorological data at the target time granularity. This multimodal time-series feature data is used to train a forecasting model for power grid load forecasting. This ensures the temporal consistency of the multimodal time-series data while preserving the original temporal characteristics of the data with larger fluctuations as much as possible through the time-series alignment operation, thereby reducing further spurious fluctuations and errors introduced by subsequent interpolation or completion, and ultimately improving the accuracy of the trained forecasting model. Attached Figure Description

[0028] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0029] Figure 1 This is a flowchart illustrating a multimodal time-aligned joint forecasting method for power grid meteorological loads according to an example embodiment of this application;

[0030] Figure 2 This is a flowchart illustrating a multimodal time-aligned joint forecasting method for power grid meteorological loads according to another exemplary embodiment of this application;

[0031] Figure 3 This is a schematic diagram of the structure of a multimodal time-aligned power grid meteorological load joint forecasting device according to an example embodiment of this application.

[0032] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0033] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0034] Figure 1 This is a flowchart illustrating a multimodal time-aligned joint forecasting method for power grid meteorological load according to an example embodiment of this application. Figure 1 As shown in this embodiment, the multimodal time-series aligned joint forecasting method for power grid meteorological load includes:

[0035] S101. Obtain power grid load data and power grid meteorological data.

[0036] First, it is necessary to obtain power grid load data and power grid meteorological data from the power grid monitoring system and meteorological monitoring stations, respectively. Among them, power grid load data is usually recorded at a certain sampling frequency (such as per minute or per hour) to record the power load of each node in the power grid. For example, the data content includes load-related parameters such as power values, frequency, and voltage of different nodes.

[0037] Meteorological data for the power grid can be obtained through meteorological monitoring stations or integrated environmental sensor clusters on the power grid. This data can include various meteorological elements such as temperature, humidity, wind speed, and precipitation, and may be recorded at different sampling frequencies. It is worth noting that if the data is obtained through an integrated environmental sensor cluster on the power grid, the sampling frequency can be set according to actual data requirements.

[0038] S102. Perform time-series alignment operation on the power grid load data and power grid meteorological data at the same target time granularity.

[0039] In this step, time-series alignment operations can be performed on the power grid load data and power grid meteorological data at the same target time granularity. The time-series alignment operation is used to align the time granularity of the data with smaller data fluctuations in the power grid load data and power grid meteorological data towards the time granularity of the data with larger data fluctuations.

[0040] It is worth noting that the core of this step is to set the target time granularity mentioned above so that when performing time-series alignment operations on power grid load data and power grid meteorological data, the time granularity of the data with smaller fluctuations in power grid load data and power grid meteorological data can be aligned with the time granularity of the data with larger fluctuations.

[0041] Furthermore, it is worth emphasizing that the aforementioned level of data fluctuation is not the same as the data sampling frequency. That is, a high sampling rate does not necessarily mean a high level of data fluctuation, and a low sampling rate does not necessarily mean a low level of data fluctuation.

[0042] Optionally, the time difference between the time granularity corresponding to the data with a large degree of fluctuation in power grid load data and power grid meteorological data and the target time granularity is smaller than the time difference between the time granularity corresponding to the data with a small degree of fluctuation and the target time granularity.

[0043] In other words, let's say the target time granularity is set to A, the time granularity of the grid load data is B, and the time granularity of the grid meteorological data is C. If it's determined that the grid load data fluctuates more than the grid meteorological data, then the target time granularity A will be closer to time granularity B within the range of B to C; that is, the distance between target time granularity A and time granularity B will be less than the distance between target time granularity A and time granularity C. Conversely, if it's determined that the grid meteorological data fluctuates more than the grid load data, then the target time granularity A will be closer to time granularity C within the range of B to C; that is, the distance between target time granularity A and time granularity C will be less than the distance between target time granularity A and time granularity B.

[0044] S103. Generate multimodal time series feature data based on the grid load data aligned at the target time granularity and the grid meteorological data, so as to use the multimodal time series feature data to train the prediction model used for grid load forecasting.

[0045] The aligned grid load data and grid meteorological data at the target time granularity will generate multimodal time-series feature data. The fusion method can be selected according to actual needs, such as simple feature concatenation, weighted summation, or using deep learning models (such as attention mechanisms) to achieve feature interaction and fusion.

[0046] For example, the extracted power grid load features and meteorological features can be spliced ​​together according to timestamps to form a multimodal time series feature vector every half hour; or an attention mechanism model can be used to weight and fuse load features and meteorological features to highlight the impact of key features on load forecasting.

[0047] Next, an LSTM neural network can be selected as the power grid load prediction model to capture long-term dependencies in time-series data. The prediction model is then trained using multimodal time-series feature data as input and actual power grid load values ​​as output. For example, the multimodal time-series feature data can be divided into training and validation sets. The LSTM model can be trained using the training set, and the model's performance can be monitored on the validation set. Training can be stopped early when performance no longer improves.

[0048] In this embodiment, grid load data and grid meteorological data are acquired. Then, at the same target time granularity, a time-series alignment operation is performed on the grid load data and grid meteorological data. The time-series alignment operation is used to align the time granularity of the data with smaller fluctuations in the grid load data and grid meteorological data towards the time granularity of the data with larger fluctuations. Then, multimodal time-series feature data is generated based on the grid load data and grid meteorological data aligned at the target time granularity. The prediction model for grid load forecasting is trained using the multimodal time-series feature data. This ensures the time consistency of the multimodal time-series data while the time-series alignment operation preserves the original time characteristics of the data with larger fluctuations as much as possible, thereby reducing further spurious fluctuations and errors introduced by subsequent interpolation or completion, and thus improving the accuracy of the trained prediction model.

[0049] Specifically, in the field of load forecasting, power grid load data and meteorological data are multimodal time-series data, requiring alignment with a unified time granularity for input into the forecasting model for training. Existing technologies typically employ either a fixed-granularity unification method or a high-frequency alignment priority method. The fixed-granularity unification method selects a fixed, unified time granularity to resample (interpolate or downsample) all data. The high-frequency alignment priority method, on the other hand, interpolates low-sampling-frequency data to the time granularity of high-sampling-frequency data, ensuring consistent time indices.

[0050] However, when highly volatile data (such as grid load) is interpolated to complete the data, it introduces unreal fluctuation peaks or valleys into the originally highly volatile time series, resulting in noise amplification. In other words, unnecessary sampling and adjustment of highly volatile data will destroy its original time series fluctuation characteristics, making it difficult for the prediction model to accurately capture the real fluctuation pattern.

[0051] The aforementioned scheme, when aligning to the same target time granularity, retains the original time granularity of data with higher volatility, and only aligns the time granularity of data with lower volatility towards the direction of data with higher volatility. This avoids the secondary error amplification caused by interpolation or completion on highly volatile data. In other words, highly volatile data, while retaining its original granularity, will not be affected by spurious fluctuations introduced by interpolation or completion, ensuring data authenticity. Consequently, during model training, the fluctuation patterns of the input features are more realistic, reducing noise interference and making the prediction results closer to actual operating conditions in terms of both trend and detail.

[0052] Figure 2 This is a flowchart illustrating a multimodal time-aligned joint forecasting method for power grid meteorological load according to another example embodiment of this application. Figure 2 As shown in this embodiment, the multimodal time-series aligned joint forecasting method for power grid meteorological load includes:

[0053] S201. Obtain power grid load data and power grid meteorological data.

[0054] This could involve acquiring power grid load data at a first sampling frequency and power grid meteorological data at a second sampling frequency, wherein the first sampling frequency and the second sampling frequency are different.

[0055] S202. Determine the target time granularity based on power grid load data and power grid meteorological data.

[0056] In this step, the target time granularity can be determined based on the power grid load data and the power grid meteorological data. The target time granularity is the time granularity corresponding to the data with large fluctuations in the power grid load data and the power grid meteorological data.

[0057] In one specific implementation, load fluctuation characteristic parameters, used to characterize the degree of load fluctuation, can be determined based on grid load data, and meteorological fluctuation characteristic parameters, used to characterize the degree of meteorological fluctuation, can be determined based on grid meteorological data. Then, the sampling rate corresponding to the larger fluctuation characteristic parameter among the load fluctuation characteristic parameters and meteorological fluctuation characteristic parameters is determined to establish a target time granularity, and the data corresponding to the smaller fluctuation parameter is aligned according to the target time granularity.

[0058] It is worth noting that the above steps do not arbitrarily select the target time granularity when determining it. Instead, the time granularity corresponding to the data with a large degree of fluctuation is used as the target. This ensures that the subsequent time alignment operation only adjusts the granularity of the data with a small degree of fluctuation, while the data with high fluctuation retains its original sampling frequency, thus avoiding the introduction of larger errors.

[0059] Optionally, the aforementioned load fluctuation characteristic parameters and / or meteorological fluctuation characteristic parameters are determined based on the ratio between at least one of the following indicators: data variance, standard deviation, or mean absolute difference, and the corresponding preset indicator. The preset indicator values ​​can be pre-set according to the system type, data source characteristics, and forecasting task requirements. For example, the reference value for the variance of the power grid load can be defined as 500 (in megawatts²), and the reference value for the standard deviation of meteorological data can be defined as 2 (in degrees Celsius²).

[0060] Optionally, in addition to the original variance, root mean square error, and mean absolute difference, the frequency domain power spectral density energy distribution ratio, the proportion of high-frequency energy in short-time Fourier transform, wavelet energy entropy, kurtosis-skewness, and density of change points can be superimposed to form a weighted comprehensive fluctuation score. Then, this comprehensive fluctuation score is used to determine the degree of data fluctuation.

[0061] S203. Perform time-series alignment operation on only the power grid load data and power grid meteorological data with small data fluctuations to the target time granularity.

[0062] If the original time granularity of the data corresponding to the smaller fluctuation parameter in the load fluctuation characteristic parameter and the meteorological fluctuation characteristic parameter is smaller than the target time granularity, then the corresponding data will be downsampled according to the target time granularity.

[0063] If the original time granularity of the data corresponding to the smaller fluctuation parameter in the load fluctuation characteristic parameter and the meteorological fluctuation characteristic parameter is larger than the target time granularity, then the corresponding data will be interpolated and completed according to the target time granularity.

[0064] It's worth noting that interpolating highly volatile data can generate inaccurate peaks or troughs in the original sequence, and may even alter local fluctuation trends. Conversely, downsampling highly volatile data can smooth out existing abrupt changes and short-period fluctuations, leading to feature loss.

[0065] Therefore, the above steps, during time alignment, preserve the original time granularity of data with high volatility, without interpolation or downsampling, ensuring that every sampling point of the data is an actual collected value. This means that the shape of the data curve is consistent with the actual operating conditions, and the volatility pattern will not be changed by the alignment operation. This ensures that the feature distribution of the model training data is consistent with the true feature distribution, avoiding misleading the model's learning. It is evident that the true volatility details of highly volatile data are fully preserved, preventing data distortion at the source and significantly improving the fidelity of training samples.

[0066] Furthermore, the above steps select the time granularity of the data with greater fluctuations as the target granularity based on the fluctuation characteristics, and only perform interpolation or downsampling on the low-fluctuation data, so that the high-fluctuation data remains unchanged at the original sampling frequency. Only one time alignment operation is required, which can effectively reduce the amount of computation when performing time alignment operation to generate training samples while ensuring the fidelity of training samples.

[0067] Furthermore, in one specific implementation, the power grid load data and power grid meteorological data can be segmented according to time periods to form power grid load data sequences and power grid meteorological data sequences. Then, within the same time period, the data with greater fluctuations in the target power grid load data sequence and the target power grid meteorological data sequence are identified. Next, within each time period, the time granularity of the data with smaller fluctuations in the target power grid load data sequence and the target power grid meteorological data sequence is aligned towards the time granularity of the data with larger fluctuations, thus forming aligned multimodal time-series feature data for each time period.

[0068] By dividing a continuous time series into multiple finite-length sub-segments, local fitting errors caused by resampling or interpolation are confined within the boundaries of these sub-segments, preventing the cumulative propagation of errors and phase drift along the entire time axis, and reducing the contamination of characteristic statistics (mean, variance, spectral energy) by the boundary segments. Furthermore, the statistical characteristics (mean drift, variance level, spectral density shape) within each segmented time period more closely resemble a locally stationary process.

[0069] Furthermore, calculating the fluctuation metrics of the load and meteorological sequences within each segment and dynamically selecting high-fluctuation data sources as granular anchors for that segment effectively avoids over-smoothing or erroneous interpolation of local high-frequency details by a globally uniform strategy. Fixing high-fluctuation sources as reference time axes without downsampling / interpolation avoids performing resampling operations with significant low-pass effects on them, preserving the phase and amplitude information of spikes, abrupt changes, and short-period components. Moreover, not interpolating high-fluctuation sources directly avoids Runge phenomenon and ringing effects generated by higher-order interpolation kernels in high-curvature regions, reducing erroneous excitation for subsequent model training.

[0070] S204. Generate multimodal time series feature data based on the grid load data aligned at the target time granularity and the grid meteorological data, so as to train the prediction model used for grid load forecasting using the multimodal time series feature data.

[0071] In this step, multimodal time-series feature data will be generated from the aligned grid load data and grid meteorological data at the target time granularity. The fusion method can be selected according to actual needs, such as simple feature concatenation, weighted summation, or using deep learning models (such as attention mechanisms) to achieve feature interaction and fusion.

[0072] Figure 3 This is a schematic diagram of the structure of a multimodal time-aligned power grid meteorological load joint forecasting device according to an example embodiment of this application. Figure 3 As shown, the multimodal time-aligned power grid meteorological load joint forecasting device 300 provided in this embodiment includes:

[0073] The acquisition module 310 is used to acquire power grid load data and power grid meteorological data;

[0074] The processing module 320 is used to perform a time-series alignment operation on the power grid load data and the power grid meteorological data at the same target time granularity. The time-series alignment operation is used to align the time granularity of the data with smaller data fluctuation in the power grid load data and the power grid meteorological data to the time granularity of the data with larger data fluctuation.

[0075] The training module 330 is used to generate multimodal time series feature data based on the aligned power grid load data and power grid meteorological data at the target time granularity, so as to train the prediction model for power grid load forecasting using the multimodal time series feature data.

[0076] Optionally, the time difference between the time granularity corresponding to the data with large fluctuations in the power grid load data and the target time granularity is smaller than the time difference between the time granularity corresponding to the data with small fluctuations and the target time granularity.

[0077] Optionally, the processing module 320 is specifically used for:

[0078] The target time granularity is determined based on the power grid load data and the power grid meteorological data. The target time granularity is the time granularity corresponding to the data with large fluctuations in the power grid load data and the power grid meteorological data.

[0079] Only data with relatively small fluctuations in the power grid load data and the power grid meteorological data are time-aligned to the target time granularity.

[0080] Optionally, the processing module 320 is specifically used for:

[0081] Based on the power grid load data, load fluctuation characteristic parameters for characterizing the degree of load fluctuation are determined, and based on the power grid meteorological data, meteorological fluctuation characteristic parameters for characterizing the degree of meteorological fluctuation are determined.

[0082] The target time granularity is determined by identifying the sampling rate corresponding to the larger fluctuation characteristic parameter among the load fluctuation characteristic parameters and the meteorological fluctuation characteristic parameters, and the data corresponding to the smaller fluctuation parameter is aligned according to the target time granularity.

[0083] Optionally, the processing module 320 is specifically used for:

[0084] If the original time granularity of the data corresponding to the smaller fluctuation parameter among the load fluctuation characteristic parameters and the meteorological fluctuation characteristic parameters is smaller than the target time granularity, then the corresponding data is downsampled according to the target time granularity.

[0085] If the original time granularity of the data corresponding to the smaller fluctuation parameter among the load fluctuation characteristic parameters and the meteorological fluctuation characteristic parameters is greater than the target time granularity, then the corresponding data is interpolated and completed according to the target time granularity.

[0086] Optionally, the load fluctuation characteristic parameter and / or the meteorological fluctuation characteristic parameter are determined based on the ratio between at least one of the data variance, root mean square error, or mean absolute difference and the corresponding preset indicator.

[0087] Optionally, the acquisition module 310 is specifically used for:

[0088] The power grid load data at a first sampling frequency and the power grid meteorological data at a second sampling frequency are acquired, wherein the first sampling frequency and the second sampling frequency are different.

[0089] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

[0090] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A multimodal time-series aligned joint forecasting method for power grid meteorological load, characterized in that, include: Acquire power grid load data and power grid meteorological data; At the same target time granularity, a time-series alignment operation is performed on the power grid load data and the power grid meteorological data. This time-series alignment operation aligns the time granularity of data with smaller fluctuations in the power grid load data and the power grid meteorological data towards a direction closer to the time granularity of data with larger fluctuations. This includes: The target time granularity is determined based on the power grid load data and the power grid meteorological data. The target time granularity is the time granularity corresponding to the data with large fluctuations in the power grid load data and the power grid meteorological data. Only data with relatively small fluctuations in the power grid load data and the power grid meteorological data are time-aligned to the target time granularity. Furthermore, during the time alignment process, no interpolation or downsampling is performed on data with large fluctuations to ensure that every sampling point of data with large fluctuations is an actual sampling point. Multimodal time-series feature data is generated based on the grid load data aligned at the target time granularity and the grid meteorological data, and the prediction model for grid load forecasting is trained using the multimodal time-series feature data.

2. The multimodal time-aligned joint forecasting method for power grid meteorological load according to claim 1, characterized in that, The time difference between the time granularity corresponding to the data with large fluctuations in the power grid load data and the target time granularity is smaller than the time difference between the time granularity corresponding to the data with small fluctuations and the target time granularity.

3. The multimodal time-aligned joint forecasting method for power grid meteorological load according to claim 1, characterized in that, Determining the target time granularity based on the power grid load data and the power grid meteorological data includes: Based on the power grid load data, load fluctuation characteristic parameters for characterizing the degree of load fluctuation are determined, and based on the power grid meteorological data, meteorological fluctuation characteristic parameters for characterizing the degree of meteorological fluctuation are determined. The target time granularity is determined by identifying the sampling rate corresponding to the larger fluctuation characteristic parameter among the load fluctuation characteristic parameters and the meteorological fluctuation characteristic parameters, and the data corresponding to the smaller fluctuation parameter is aligned according to the target time granularity.

4. The multimodal time-aligned joint forecasting method for power grid meteorological load according to claim 3, characterized in that, The step of aligning the data corresponding to another smaller fluctuation parameter according to the target time granularity includes: If the original time granularity of the data corresponding to the smaller fluctuation parameter among the load fluctuation characteristic parameters and the meteorological fluctuation characteristic parameters is smaller than the target time granularity, then the corresponding data is downsampled according to the target time granularity. If the original time granularity of the data corresponding to the smaller fluctuation parameter among the load fluctuation characteristic parameters and the meteorological fluctuation characteristic parameters is greater than the target time granularity, then the corresponding data is interpolated and completed according to the target time granularity.

5. The multimodal time-aligned joint forecasting method for power grid meteorological load according to claim 3, characterized in that, The load fluctuation characteristic parameters and / or the meteorological fluctuation characteristic parameters are determined based on the ratio between at least one of the data variance, root mean square error, or mean absolute difference and the corresponding preset index.

6. The multimodal time-aligned joint forecasting method for power grid meteorological load according to claim 1, characterized in that, The acquisition of power grid load data and power grid meteorological data includes: The power grid load data at a first sampling frequency and the power grid meteorological data at a second sampling frequency are acquired, wherein the first sampling frequency and the second sampling frequency are different.

7. A multimodal time-aligned power grid meteorological load joint forecasting device, characterized in that, include: The acquisition module is used to acquire power grid load data and power grid meteorological data; The processing module is used to perform time-series alignment operations on the power grid load data and the power grid meteorological data at the same target time granularity. The time-series alignment operation is used to align the time granularity of the data with smaller data fluctuations in the power grid load data and the power grid meteorological data to the time granularity of the data with larger data fluctuations. The processing module is specifically used for: The target time granularity is determined based on the power grid load data and the power grid meteorological data. The target time granularity is the time granularity corresponding to the data with large fluctuations in the power grid load data and the power grid meteorological data. Only data with relatively small fluctuations in the power grid load data and the power grid meteorological data are time-aligned to the target time granularity. Furthermore, during the time alignment process, no interpolation or downsampling is performed on data with large fluctuations to ensure that every sampling point of data with large fluctuations is an actual sampling point. The training module is used to generate multimodal time-series feature data based on the aligned grid load data and grid meteorological data at the target time granularity, so as to train the prediction model for grid load forecasting using the multimodal time-series feature data.

8. The multimodal time-aligned power grid meteorological load joint forecasting device according to claim 7, characterized in that, The time difference between the time granularity corresponding to the data with large fluctuations in the power grid load data and the target time granularity is smaller than the time difference between the time granularity corresponding to the data with small fluctuations and the target time granularity.

Images