Meteorological data fusion method and device, computer equipment, storage medium and product

Abstract

The application relates to a meteorological data fusion method and device, computer equipment, a storage medium and a product. The method comprises the following steps: acquiring measured meteorological data and predicted meteorological data of a plurality of meteorological data sources at a current time; performing weighted fusion on the measured meteorological data and the predicted meteorological data of the plurality of meteorological data sources according to a first weighting parameter corresponding to a target user to obtain a first meteorological feature vector; and performing weighted fusion on the first meteorological feature vector and a second meteorological feature vector of at least one other user according to a second weighting parameter of the target user and the at least one other user to obtain a target meteorological feature vector corresponding to the target user; wherein the second meteorological feature vector is obtained by performing weighted fusion on the measured meteorological data and the predicted meteorological data of the plurality of meteorological data sources based on a weighting parameter corresponding to the other user. By using the method, the fusion effect of the meteorological data can be improved, and the precision and accuracy of the power consumption prediction can be improved.

Description

Technical Field

[0001] This application relates to the field of data fusion technology, and in particular to a meteorological data fusion method, apparatus, computer equipment, storage medium and product. Background Technology

[0002] In user electricity consumption forecasting scenarios, it is usually necessary to refer to current weather data, and high-quality weather data can greatly increase the accuracy of user electricity consumption forecasts. Weather data can be divided into weather forecast data and actual weather data. Both forecast and actual data may have multiple available data sources, and there are differences between different data sources. It is necessary to fuse forecast and actual data from multiple data sources to improve the quality of weather data.

[0003] Traditional meteorological data fusion methods suffer from poor fusion results, leading to low accuracy and poor precision in electricity consumption forecasting. Summary of the Invention

[0004] Therefore, it is necessary to provide a meteorological data fusion method, device, computer equipment, computer-readable storage medium, and computer program product that can improve the fusion effect of meteorological data and thus improve the accuracy and precision of electricity consumption forecasting, in order to address the above-mentioned technical problems.

[0005] Firstly, this application provides a meteorological data fusion method, including:

[0006] Acquire measured and predicted meteorological data from multiple meteorological data sources at the current moment;

[0007] Based on the first weighting parameter corresponding to the target user, the measured meteorological data and predicted meteorological data from multiple meteorological data sources are weighted and fused to obtain the first meteorological feature vector;

[0008] Based on the second weighting parameters of the target user and at least one other user, the first meteorological feature vector and the second meteorological feature vector of at least one other user are weighted and fused to obtain the target meteorological feature vector corresponding to the target user; wherein, the second meteorological feature vector is obtained by weighting and fusing measured meteorological data and predicted meteorological data from multiple meteorological data sources based on the weighting parameters corresponding to other users.

[0009] In one embodiment, the first weighting parameter includes a data weighting parameter and a data source weighting parameter. Based on the first weighting parameter corresponding to the target user, measured meteorological data and predicted meteorological data from multiple meteorological data sources are weighted and fused to obtain a first meteorological feature vector, including:

[0010] For each meteorological data source, the measured meteorological data and the predicted meteorological data of the meteorological data source are weighted and fused according to the data weighting parameters to obtain the third meteorological feature vector of the meteorological data source;

[0011] Based on the weighting parameters of the data sources, the third meteorological feature vectors of each meteorological data source are weighted and fused to obtain the first meteorological feature vector.

[0012] In one embodiment, the method further includes:

[0013] Acquire sample meteorological data from multiple users; the sample meteorological data includes measured data and predicted data from multiple meteorological data sources at multiple times.

[0014] For each user, the user's sample meteorological data is input into a preset feature extraction model to determine the measured feature vector and predicted feature vector of each meteorological data source of the user at each time.

[0015] The measured and predicted feature vectors of each meteorological data source for each user at each time point are input into the initial meteorological fusion network to obtain the sample fusion feature vector at each time point; the initial meteorological fusion network is a fusion network based on a hybrid attention mechanism.

[0016] Based on the sample fusion feature vectors at each time point, the historical electricity consumption of the target user at each time point, and the preset electricity consumption prediction model, the initial meteorological fusion network is iteratively trained to obtain the target meteorological fusion model corresponding to the target user; the target meteorological fusion model includes the first weighting parameter and the second weighting parameter.

[0017] In one embodiment, the user's sample meteorological data is input into a preset feature extraction model to determine the measured feature vectors and predicted feature vectors of each meteorological data source at each time point, including:

[0018] Determine the basic characteristics of the sample meteorological data; the basic characteristics include at least one of the meteorological value at the current moment, the average meteorological value over a preset time period prior to the current moment, and the meteorological rate of change.

[0019] The sample meteorological data and basic data features are input into the preset feature extraction model to determine the measured feature vector and predicted feature vector of each meteorological data source of the user at each time.

[0020] In one embodiment, the initial meteorological fusion network includes a first fusion network, a second fusion network, and a third fusion network. The measured and predicted feature vectors of each user's meteorological data sources at each time point are input into the initial meteorological fusion network to obtain the sample fusion feature vector at each time point, including:

[0021] For each user's meteorological data source, the measured feature vector and predicted feature vector at each time moment are input into the first fusion network to obtain the first fusion feature vector at each time moment;

[0022] For each time point, the first fusion feature vector of each meteorological data source of the user is input into the second fusion network to obtain the second fusion feature vector of each time point;

[0023] The second fusion feature vector of each user is input into the third fusion network to obtain the sample fusion feature vector at each time step.

[0024] In one embodiment, based on the sample fusion feature vectors at each time point, the target user's historical electricity consumption at each time point, and a preset electricity consumption prediction model, the initial meteorological fusion network is iteratively trained to obtain the target meteorological fusion model corresponding to the target user, including:

[0025] Based on the target user’s historical electricity consumption at each time point, determine the historical electricity consumption sequence corresponding to each time point;

[0026] The sample fusion feature vectors and historical electricity consumption sequences at each time point are input into the preset electricity consumption prediction model to obtain the predicted electricity consumption at each time point.

[0027] Based on the error between the predicted electricity consumption at each time point and the historical electricity consumption at each time point, the initial meteorological fusion network is iteratively trained to obtain the target meteorological fusion model corresponding to the target user.

[0028] In one embodiment, sample meteorological data from multiple users is acquired, including:

[0029] Acquire initial meteorological data from multiple users; the initial meteorological data includes initial measured data and initial forecast data from multiple meteorological data sources at multiple times;

[0030] The initial meteorological data is preprocessed to obtain sample meteorological data for each user; the preprocessing includes at least one of time alignment, spatial alignment and standardization.

[0031] In one embodiment, the method further includes:

[0032] Obtain the target user's historical electricity consumption up to the current moment;

[0033] The target meteorological feature vector and historical electricity consumption are input into the preset electricity consumption prediction model to determine the predicted electricity consumption of the target user.

[0034] Secondly, this application also provides a meteorological data fusion device, comprising:

[0035] The first acquisition module is used to acquire measured meteorological data and forecast meteorological data from multiple meteorological data sources at the current moment;

[0036] The first fusion module is used to perform weighted fusion of measured meteorological data and predicted meteorological data from multiple meteorological data sources according to the first weighting parameter corresponding to the target user, so as to obtain the first meteorological feature vector.

[0037] The second fusion module is used to perform weighted fusion of the first meteorological feature vector and the second meteorological feature vector of at least one other user based on the second weighting parameters of the target user and at least one other user, to obtain the target meteorological feature vector corresponding to the target user; wherein, the second meteorological feature vector is obtained by weighted fusion of measured meteorological data and predicted meteorological data from multiple meteorological data sources based on the weighting parameters corresponding to other users.

[0038] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the meteorological data fusion method in the first aspect described above.

[0039] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the meteorological data fusion method described in the first aspect above.

[0040] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the meteorological data fusion method described in the first aspect above.

[0041] The aforementioned meteorological data fusion method, apparatus, computer equipment, storage medium, and computer program product acquire measured meteorological data and forecast meteorological data from multiple meteorological data sources at the current moment; based on a first weighting parameter corresponding to the target user, the measured meteorological data and forecast meteorological data from multiple meteorological data sources are weighted and fused to obtain a first meteorological feature vector; based on a second weighting parameter between the target user and at least one other user, the first meteorological feature vector and the second meteorological feature vector of at least one other user are weighted and fused to obtain a target meteorological feature vector corresponding to the target user; wherein, the second meteorological feature vector is obtained by weighting and fusing the measured meteorological data and forecast meteorological data from multiple meteorological data sources based on the weighting parameters corresponding to other users. In other words, when performing meteorological data fusion in this application embodiment, it comprehensively considers both measured and predicted data, as well as different meteorological data sources, and even incorporates the meteorological characteristics of other users. That is, it performs comprehensive fusion of meteorological data from the dimensions of data type, data source, and user. Furthermore, in this application solution, a set of meteorological fusion parameters suitable for different users is learned to obtain target meteorological feature vectors applicable to the target users, which can improve the fusion effect of meteorological data. Consequently, when predicting electricity consumption based on the target meteorological feature vectors of the target users obtained after fusion, the accuracy of electricity consumption prediction for the target users can be greatly improved, thus enhancing the accuracy of predicted electricity consumption. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is an application environment diagram of the meteorological data fusion method in one embodiment;

[0044] Figure 2 This is a flowchart illustrating a meteorological data fusion method in one embodiment;

[0045] Figure 3 This is a flowchart illustrating the meteorological data fusion method in another embodiment;

[0046] Figure 4 This is a schematic diagram of cross-attention fusion and power consumption prediction in one embodiment;

[0047] Figure 5 This is a flowchart illustrating the meteorological data fusion method in another embodiment;

[0048] Figure 6 This is a structural block diagram of a meteorological data fusion device in one embodiment;

[0049] Figure 7 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0051] User electricity consumption prediction scenarios require high-quality weather data. By combining historical electricity consumption sequences with corresponding weather conditions, the correlation between user electricity consumption habits and weather can be characterized, thereby improving the model's accuracy in predicting electricity consumption.

[0052] Weather data is divided into forecast data and measured data. Both forecast and measured data may have multiple available data sources, such as weather station A / B databases, satellite remote sensing, and third-party APIs. Differences exist between these data sources, necessitating the fusion of forecast and measured data from multiple sources to improve the quality of weather data. However, existing meteorological data fusion methods suffer from poor fusion results, leading to low accuracy and precision in electricity consumption forecasts.

[0053] Based on this, this application provides a meteorological data fusion method. By constructing a multi-layer attention mechanism and learning a set of weather fusion weighting parameters for each user, comprehensive weather characteristics are constructed for each user and entered into the user's electricity consumption prediction model. This not only improves the fusion effect of meteorological data, but also improves the accuracy and precision of electricity consumption prediction.

[0054] The meteorological data fusion method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, different meteorological data sources 101 communicate with server 102 via a network. A data storage system can store the data that server 102 needs to process. The data storage system can be integrated onto server 102, or it can be located in the cloud or on other network servers. The meteorological data sources 101 can be, but are not limited to, weather station A / B, satellite remote sensing, third-party APIs, etc. Server 102 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0055] In one exemplary embodiment, such as Figure 2 As shown, a meteorological data fusion method is provided, which can be applied to... Figure 1Taking the server in the example, the explanation includes the following steps 201 to 203. Wherein:

[0056] Step 201: Obtain measured meteorological data and predicted meteorological data from multiple meteorological data sources at the current time.

[0057] Meteorological data sources can include multiple different types of meteorological data sources, such as weather stations, satellite remote sensing, and third-party sources. Each type of meteorological data source can also include multiple data sources in different locations and regions, such as setting up multiple weather stations in different areas of a city, such as weather station A, weather station B, and weather station C. In this example, multiple meteorological data sources can cover official meteorological department monitoring sites (such as national meteorological observation stations and regional automatic weather stations), professional meteorological observation equipment networks (such as ground observation stations, upper-air sounding stations, radar observation stations, and satellite remote sensing systems), monitoring nodes of third-party meteorological service agencies, industry-specific meteorological monitoring equipment (such as customized observation equipment for agriculture, transportation, and aviation), and distributed civilian meteorological observation terminals, ensuring the diversity and coverage of data sources and avoiding problems such as monitoring blind spots and data bias that may exist with a single data source.

[0058] When performing meteorological data fusion, it is necessary to integrate meteorological data from multiple meteorological data sources to improve the quality of the meteorological data. For example, these multiple meteorological data sources may include multiple meteorological data sources of different types, and / or multiple meteorological data sources at different locations within the same type. For each meteorological data source, the measured meteorological data and predicted meteorological data corresponding to the current moment can be obtained. The measured meteorological data corresponding to the current moment may include meteorological data measured in real time at the current moment, as well as meteorological data actually measured within a preset time period prior to the current moment. The predicted meteorological data corresponding to the current moment may include meteorological data for at least one time period after the current moment, predicted based on the measured meteorological data within the preset time period prior to the current moment.

[0059] For example, when acquiring meteorological data from various meteorological data sources, real-time data capture and synchronous updates of each meteorological data source can be achieved through standardized interfaces and data transmission protocols (such as HTTP, MQTT, FTP, etc.). At the same time, preprocessing operations such as integrity verification, outlier removal, and accuracy calibration can be performed on the acquired measured meteorological data.

[0060] Step 202: Based on the first weighting parameter corresponding to the target user, the measured meteorological data and predicted meteorological data from multiple meteorological data sources are weighted and fused to obtain the first meteorological feature vector.

[0061] In order to adapt to the differences and personalization of users, a set of meteorological fusion parameters is learned for different users. The meteorological fusion parameters may include the first weighted parameter here and the second weighted parameter in the following steps. For example, a set of meteorological fusion parameters can be learned for each user during the training and learning phase; or, the user meteorological fusion parameters can be continuously updated based on real-time data through continuous learning.

[0062] Assuming that the meteorological fusion parameters corresponding to the target user have been learned before the current moment, including the first weighting parameter, when performing multi-source meteorological data fusion, the measured meteorological data and predicted meteorological data from multiple meteorological data sources can be weighted and fused based on the first weighting parameter to obtain the first meteorological feature vector.

[0063] For example, the first weighting parameter may include a data weighting parameter and a data source weighting parameter. When performing weighted fusion of measured meteorological data and predicted meteorological data from multiple meteorological data sources, the measured meteorological data and predicted meteorological data from each meteorological data source can first be weighted and fused according to the data weighting parameter to obtain the third meteorological feature vector of that meteorological data source. Then, the third meteorological feature vectors of each meteorological data source can be weighted and fused according to the data source weighting parameter to obtain the first meteorological feature vector. In other words, the measured meteorological data and predicted meteorological data can first be weighted and fused on a data source basis to obtain the third meteorological feature vector of each meteorological data source. Then, the different meteorological data sources can be weighted and fused, that is, the third meteorological feature vectors of each meteorological data source can be weighted and fused according to the data source weighting parameter to obtain the first meteorological feature vector after fusing the measured meteorological data, predicted meteorological data, and different meteorological data sources.

[0064] For example, when performing weighted fusion of meteorological data based on data weighting parameters and data source weighting parameters, one can first perform weighted fusion of measured meteorological data from each meteorological data source based on data source weighting parameters to obtain a measured meteorological feature vector, and then perform weighted fusion of predicted meteorological data from each meteorological data source based on data source weighting parameters to obtain a predicted meteorological feature vector; then, perform weighted fusion of the measured meteorological feature vector and the predicted meteorological feature vector based on data source weighting parameters to obtain a first meteorological feature vector that integrates measured meteorological data, predicted meteorological data, and different meteorological data sources.

[0065] It should be noted that the first meteorological feature vector is a high-dimensional vector that integrates multiple data sources and multiple meteorological quantities. For example, it includes multiple meteorological quantities from multiple data sources, including but not limited to temperature, humidity, surface shortwave radiation, 10-meter wind speed, and 10-meter wind direction. Different users will have personalized meteorological data weighting preferences; that is, the data weighting parameters and data source weighting parameters are different for different users. Based on the weighting parameters of different users, multiple meteorological quantities from multiple data sources can be weighted and fused to obtain the final fused first meteorological feature vector. Before training, each user's initial weights are the same, i.e., equal-weight initialization. In the long-interval electricity consumption prediction training, the equal-weight initialization is iteratively trained to obtain each user's different weight preferences.

[0066] Step 203: Based on the second weighting parameter of the target user and at least one other user, the first meteorological feature vector and the second meteorological feature vector of at least one other user are weighted and fused to obtain the target meteorological feature vector corresponding to the target user.

[0067] The second meteorological feature vector is obtained by weighting and fusing measured meteorological data and predicted meteorological data from multiple meteorological data sources based on weighting parameters corresponding to other users. The weighting parameters corresponding to other users are similar to the meteorological fusion parameters corresponding to the target user mentioned above, and may also include data weighting parameters and data source weighting parameters. Based on the data weighting parameters and data source weighting parameters corresponding to other users, measured meteorological data and predicted meteorological data from multiple meteorological data sources at the current time can be weighted and fused to obtain the second meteorological feature vector corresponding to other users.

[0068] It should be noted that the other users corresponding to each user can be the same or different. The other users can be at least one user with a strong correlation with the target user. The correlation can be considered from at least one dimension such as electricity consumption similarity, user characteristic similarity, user relationship, user type, and regional location. For example, in the process of learning meteorological fusion parameters, at least one other user with a strong correlation with the target user and a second weighting parameter between the target user and at least one other user can be determined. Then, when fusing meteorological data, the meteorological characteristics of the target user can be corrected and updated by referring to the meteorological characteristics of the at least one other user.

[0069] For example, the meteorological fusion parameters corresponding to the target user may further include a second weighting parameter. After obtaining the first meteorological feature vector corresponding to the target user by weighting and fusing measured meteorological data and predicted meteorological data from multiple meteorological data sources based on the first weighting parameter corresponding to the target user, the first meteorological feature vector corresponding to the target user and at least one second meteorological feature vector corresponding to other users may be further weighted and fused based on the second weighting parameter between the target user and other users. That is, meteorological data fusion is performed from the user dimension to obtain the target meteorological feature vector corresponding to the target user. This target meteorological feature vector can be used to predict the electricity consumption of the target user.

[0070] In the above meteorological data fusion method, measured meteorological data and predicted meteorological data from multiple meteorological data sources are acquired at the current time. Then, based on the first weighting parameter corresponding to the target user, the measured meteorological data and predicted meteorological data from multiple meteorological data sources are weighted and fused to obtain a first meteorological feature vector. Then, based on the second weighting parameter of the target user and at least one other user, the first meteorological feature vector and the second meteorological feature vector of at least one other user are weighted and fused to obtain the target meteorological feature vector corresponding to the target user. The second meteorological feature vector is obtained by weighting and fusing the measured meteorological data and predicted meteorological data from multiple meteorological data sources based on the weighting parameter corresponding to other users. In other words, when performing meteorological data fusion in this application embodiment, it comprehensively considers both measured and predicted data, as well as different meteorological data sources, and even incorporates the meteorological characteristics of other users. That is, it performs comprehensive fusion of meteorological data from the dimensions of data type, data source, and user. Furthermore, in this application solution, a set of meteorological fusion parameters suitable for different users is learned to obtain target meteorological feature vectors applicable to the target users, which can improve the fusion effect of meteorological data. Consequently, when predicting electricity consumption based on the target meteorological feature vectors of the target users obtained after fusion, the accuracy of electricity consumption prediction for the target users can be greatly improved, thus enhancing the accuracy of predicted electricity consumption.

[0071] In one exemplary embodiment, a process for obtaining meteorological fusion parameters corresponding to a target user is provided, such as... Figure 3 As shown, the above method may further include steps 301 to 304. Wherein:

[0072] Step 301: Obtain sample meteorological data from multiple users.

[0073] The sample meteorological data includes measured data and predicted data from multiple meteorological data sources at multiple times. For example, the server can acquire meteorological data (historical observations, denoted as D_obs) from each meteorological data source in real time and store it in the data storage system. The server can also perform meteorological forecasts based on the real-time acquired measured data to obtain predicted data (future forecasts relative to historical times, denoted as D_pred), and store the predicted data in the data storage system as well. During the parameter learning phase, the server can acquire measured data and predicted data from multiple meteorological data sources at multiple historical times from the data storage system as training samples. The measured data from different meteorological data sources can be represented as S1_obs, S2_obs, ..., Sn_obs, and the predicted data from different meteorological data sources can be represented as S1_pred, S2_pred, ..., Sn_pred, where n is the number of meteorological data sources.

[0074] Since the meteorological data fusion method proposed in this application embodiment also performs weighted fusion of meteorological data from the user dimension, it is necessary to obtain sample meteorological data from multiple different users during the parameter learning process in order to learn the data correlation between different users.

[0075] In one optional implementation, the method of obtaining sample meteorological data from multiple users may further include: obtaining initial meteorological data from multiple users, wherein the initial meteorological data includes initial measured data and initial forecast data from multiple meteorological data sources at multiple times; then, preprocessing the initial meteorological data from multiple users to obtain sample meteorological data from each user; optionally, the preprocessing may include at least one of time alignment processing, spatial alignment processing, and normalization processing.

[0076] For example, time alignment can be performed by downsampling the initial measured data (which may be at the minute level) and interpolating the initial predicted data (which may be at multiple hourly intervals) based on a target time granularity (e.g., hourly). This ensures that all measured and predicted data have consistent timestamps (e.g., t=0,1,…,T). In other words, if the initial meteorological data (including initial measured and predicted data) has a small time granularity, it can be downsampled; if the initial meteorological data has a large time granularity, it can be interpolated to achieve time alignment of the meteorological data.

[0077] For example, spatial alignment can be based on a target spatial granularity (e.g., city level). If the initial meteorological data (including initial measured data and initial forecast data) is a regional (smaller than city level) mean, the initial meteorological data from multiple regions can be aggregated, such as by weighted averaging, to obtain city-level meteorological data. If the initial meteorological data is station-level (larger than city level) data, it can be converted to city-level meteorological data through interpolation to ensure that the spatial granularity of all measured and forecast data remains consistent. In other words, if the spatial granularity of the initial meteorological data is small, it can be aggregated; if the spatial granularity is large, it can be interpolated to achieve spatial alignment of the meteorological data.

[0078] For example, standardization can involve unifying the format of initial meteorological data (including initial measured data and initial forecast data). For instance, meteorological data collected and recorded from different data sources may have different formats, such as different data frame formats, data types, and data units. During comprehensive learning processing, it is necessary to unify the data format and standardize all meteorological data to prepare for subsequent data processing. For example, when standardizing data units, a normalization approach can be used, performing Z-score standardization on each meteorological data point according to its data source. ,in, This is a statistical value of historical data from this data source to avoid the impact of differences in units on the fusion.

[0079] Furthermore, it should be noted that the execution order of different preprocessing operations in the embodiments of this application is not limited. Time alignment can be performed first, followed by spatial alignment, and finally normalization. Of course, other processing orders are also possible, which will not be described in detail here.

[0080] Step 302: For each user, input the user's sample meteorological data into the preset feature extraction model to determine the measured feature vector and predicted feature vector of each meteorological data source of the user at each time.

[0081] For example, the feature extraction model can be any type of neural network model used to extract features from sample meteorological data, thereby obtaining a fixed-dimensional meteorological feature vector, including measured feature vectors and predicted feature vectors. Based on the user's sample meteorological data, sample datasets corresponding to that user at different times can be generated. For instance, a portion of sample meteorological data within a preset time period before the current time can be selected from the sample meteorological data as the sample dataset corresponding to the current time. The preset time period can be an integer multiple of the aforementioned time granularity; for example, sample meteorological data within 10 hours before the current time can be used as the sample dataset corresponding to the current time.

[0082] For example, taking a single meteorological data source for a single user as an example, for sample measured data, the measured dataset corresponding to each time moment can be determined. Then, the measured datasets at each time moment are input into a preset feature extraction model. By performing feature extraction on the measured datasets at each time moment, the measured feature vector at each time moment can be obtained. Similarly, for sample predicted data, the predicted dataset corresponding to each time moment can be determined. Then, the predicted datasets at each time moment are input into a preset feature extraction model. By performing feature extraction on the predicted datasets at each time moment, the predicted feature vector at each time moment can be obtained.

[0083] For example, when performing feature extraction, the basic data features of the sample meteorological data can also be referenced, including but not limited to at least one of the following: the meteorological value at the current moment, the average meteorological value within a preset time period (time window) before the current moment, and the meteorological rate of change. The meteorological value at the current moment can include a series of meteorological data such as temperature, humidity, surface shortwave radiation, 10-meter wind speed, and 10-meter wind direction. The meteorological average corresponds to the rolling average of meteorological values ​​over past time periods, such as the average temperature, average humidity, average shortwave radiation, average wind speed, and average wind direction within a preset time period before the current moment. The rate of change is the temporal rate of change of the meteorological values. These basic data features (including meteorological values, averages, and rates of change) are used to enhance the robustness of feature extraction. They are input together with the original sample meteorological data into the preset feature extraction model to generate richer measured feature vectors and predicted feature vectors, which helps to improve the fusion accuracy.

[0084] Based on this, we can first determine the basic characteristics of the sample meteorological data. Then, we input the sample meteorological data and the basic characteristics into a preset feature extraction model to determine the measured feature vectors and predicted feature vectors of each meteorological data source for each user at each time. Continuing with the example of a single meteorological data source for a single user, for the sample measured data, based on the measured datasets at each time, we can further determine the meteorological mean and meteorological rate of change corresponding to each time. For each time, the meteorological value at the current time, the meteorological mean at the current time, and the meteorological rate of change corresponding to the current time are used as the basic characteristics of the data at the current time. Then, during feature extraction, the measured dataset at the current time and the basic characteristics of the data at the current time can be input together into the preset feature extraction model to output the measured feature vector corresponding to the current time.

[0085] Using the same processing method, the measured feature vectors and predicted feature vectors of each meteorological data source for each user at each time can be obtained.

[0086] Step 303: Input the measured feature vectors and predicted feature vectors of each meteorological data source of each user at each time into the initial meteorological fusion network to obtain the sample fusion feature vectors at each time.

[0087] The initial meteorological fusion network is a fusion network based on a hybrid attention mechanism. Optionally, a multi-layer attention mechanism can be adopted to set up attention fusion networks for actual measurements and forecasts, multiple meteorological data sources, and multiple users, and perform weighted fusion of attention layer by layer to finally obtain the sample fusion feature vector at each time point.

[0088] For example, the initial meteorological fusion network may include a first fusion network, a second fusion network, and a third fusion network. The first fusion network is used to perform attention weighting on the measured feature vectors and predicted feature vectors; the second fusion network is used to perform attention weighting on the feature vectors from different meteorological data sources; and the third fusion network is used to perform attention weighting on the feature vectors from different users. For instance, for each user's meteorological data source, the measured and predicted feature vectors at each time point can be input into the first fusion network to obtain the first fusion feature vector at each time point. Then, for each time point, the first fusion feature vectors from each user's meteorological data source are input into the second fusion network to obtain the second fusion feature vector at each time point. Finally, the second fusion feature vectors from each user are input into the third fusion network to obtain the sample fusion feature vector at each time point.

[0089] The initial meteorological fusion network is fed into the measured and predicted feature vectors of each meteorological data source for each user at each time point. The network uses an attention mechanism to find related items across multiple dimensions, such as time series, data sources, and meteorological quantities, and then weights them. In other words, at different fusion stages, the feature vectors are fused based on attention, and the corresponding fusion network parameters are optimized iteratively.

[0090] For example, the aforementioned fusion network can employ network structures such as one-dimensional convolutional neural networks (1D-CNN) and long short-term memory networks (LSTM), extracting local / global temporal patterns and outputting fixed-dimensional vectors, such as... (d is the embedding dimension, i.e., the vector length, such as 64). Taking an LSTM network as an example, The fused feature vector of the i-th meteorological data source at time t can be represented as:

[0091]

[0092] It should be noted that the above-described initial meteorological fusion network structure is only one example. In practical applications, there can be various fusion sequences for measured and predicted data, multiple meteorological data sources, and multiple users. Correspondingly, there can also be various initial meteorological fusion network structures, which will not be elaborated on in this example.

[0093] For example, refer to Figure 4 As shown, multiple meteorological data sources can also be cross-fused. For example, the first fusion network can apply attention weighting to the measured and predicted feature vectors of the same meteorological data source, such as applying attention weighting to the predicted feature vector of predicted data source 1 and the measured feature vector of measured data source 1, where predicted data source 1 and measured data source 1 are the same meteorological data source 1. It can also apply attention weighting to the measured and predicted feature vectors of different meteorological data sources, such as applying attention weighting to the predicted feature vector of predicted data source 1 and the measured feature vector of measured data source 2, where predicted data source 1 belongs to meteorological data source 1 and measured data source 2 belongs to meteorological data source 2. Similarly, other cross-fused first fusion feature vectors can be obtained. Further, the second fusion network can fuse all the first fusion feature vectors of the user at a certain time obtained by the first fusion network to obtain the second fusion feature vector. Finally, the second fusion feature vectors of each user are input into the third fusion network to obtain the sample fusion feature vectors at each time point.

[0094] The first and second fusion networks can form a weather internal attention mechanism fusion network to fuse meteorological data. The measured data (high precision but lagging) and the predicted data (real-time but large error) are complementary. The first fusion network uses cross-attention to mine the correlation (such as the deviation of the measured temperature from the predicted temperature). The error time and feature dimensions of the predicted data from different meteorological data sources are different. Therefore, the second fusion network performs attention weighting between the first fusion feature vectors of different meteorological data sources.

[0095] In addition, users can not only use their own fused weather data, but also weight the fused weather data of different users to obtain a set of weather data that is more suitable for electricity consumption prediction. Therefore, in terms of users, after the second fusion stage, namely "weather internal attention" fusion, attention weighting is applied to the second fused feature vector of each user to obtain a sample fused feature vector that incorporates the meteorological features of other users.

[0096] Step 304: Based on the sample fusion feature vectors at each time point, the historical electricity consumption of the target user at each time point, and the preset electricity consumption prediction model, the initial meteorological fusion network is iteratively trained to obtain the target meteorological fusion model corresponding to the target user.

[0097] The target meteorological fusion model includes a first weighting parameter and a second weighting parameter. The first weighting parameter includes data weighting parameters and data source weighting parameters. That is, the target meteorological fusion model includes data weighting parameters between actual measurements and forecasts, data source weighting parameters between different meteorological data sources, and second weighting parameters between different users.

[0098] For example, the iterative training process may include: determining the historical electricity consumption sequence corresponding to each time based on the historical electricity consumption of the target user at each time; inputting the sample fusion feature vector at each time and the historical electricity consumption sequence at each time into the preset electricity consumption prediction model to obtain the predicted electricity consumption at each time; and iteratively training the initial meteorological fusion network based on the error between the predicted electricity consumption at each time and the historical electricity consumption at each time to obtain the target meteorological fusion model corresponding to the target user.

[0099] For example, the electricity consumption prediction model can be a pre-trained network model for electricity consumption prediction. During the training of the meteorological fusion model, the pre-trained electricity consumption prediction model is used to predict electricity consumption based on the sample fusion feature vector obtained from the initial meteorological fusion network. The predicted electricity consumption is obtained, and the network parameters in the initial meteorological fusion network are adjusted by comparing the differences between the historical actual electricity consumption and the predicted electricity consumption. Through iterative training, the trained target meteorological fusion model can be obtained. The network parameters in the initial meteorological fusion network can include the weighted parameters in each fusion network, namely, the weighted parameters between measured and predicted data in the first fusion network (i.e., data weighted parameters), the weighted parameters between different meteorological data sources in the second fusion network (i.e., data source weighted parameters), and the weighted parameters between different users in the third fusion network (i.e., second weighted parameters). Therefore, the target meteorological fusion model includes the trained data weighted parameters, data source weighted parameters, and second weighted parameters.

[0100] For example, the error between the predicted electricity consumption and the historical electricity consumption can be expressed as mean squared error (MSE), absolute error, cross-entropy loss, etc., and this embodiment does not specifically limit it.

[0101] It should be noted that, for different users, the meteorological fusion network is trained separately based on their historical electricity consumption at various times. This allows each user to learn a target meteorological fusion model, meaning each user learns a set of meteorological fusion parameters tailored to their specific needs. To prevent overfitting, the model training employs several methods: First, it utilizes sample meteorological data and historical electricity consumption from multiple users for joint learning. This multi-user fusion network increases data diversity and helps prevent overfitting. Second, regularization is used during training to constrain the weighting coefficients, further preventing overfitting. Third, the prediction is a time-series prediction, and the continuity of the time-series data provides a regularization effect, also preventing overfitting.

[0102] In this embodiment, a multi-layer attention mechanism fusion network is employed to fuse meteorological data from the perspectives of data type, data source, and user. Combined with historical electricity consumption data from different users, target meteorological fusion models are trained for each user separately to form meteorological fusion parameters corresponding to each user, meeting their different electricity consumption habits and needs. Specifically, the hybrid attention mechanism captures the temporal dependencies, inter-source complementarity, and dynamic weights of multi-source data to generate high-quality fusion representations. Thus, during meteorological data fusion, measured and predicted data from different meteorological data sources, as well as the meteorological characteristics of different users, can be comprehensively fused according to the meteorological fusion parameters corresponding to different users. This results in more accurate meteorological fusion features that match the user, enabling precise predictions of user electricity consumption and improving the accuracy of electricity consumption forecasts.

[0103] In one exemplary embodiment, a method for predicting electricity consumption is provided, such as... Figure 5 As shown, the above method may further include steps 501 to 502. Wherein:

[0104] Step 501: Obtain the target user's historical electricity consumption up to the current moment.

[0105] Among them, historical electricity consumption can be the actual electricity consumption of the target user at multiple historical moments before the current moment, that is, the measured electricity consumption.

[0106] For example, the server can obtain the target user's historical electricity consumption up to the current moment from the power grid system.

[0107] It should be noted that the first server for meteorological data fusion and the second server for electricity consumption forecasting can be the same server or different servers. If they are two different servers, when the second server performs electricity consumption forecasting for the target user, it can obtain the target meteorological feature vector corresponding to the target user from the first server, and at the same time obtain the target user's historical electricity consumption.

[0108] For example, the second server can send a meteorological feature acquisition instruction carrying the identity identifier (such as ID) of the target user to the first server. The first server responds to the meteorological feature acquisition instruction and, based on the identity identifier of the target user, determines the meteorological fusion parameters corresponding to the target user and the meteorological fusion parameters corresponding to at least one other user associated with the target user, including a first weighting parameter and a second weighting parameter. Then, using the method steps in the above embodiment, the first weighting parameter corresponding to the target user is used to fuse the measured meteorological data and predicted meteorological data from multiple meteorological data sources at the current time to obtain a first meteorological feature vector corresponding to the target user. The first weighting parameter corresponding to other users is used to fuse the measured meteorological data and predicted meteorological data from multiple meteorological data sources at the current time to obtain a second meteorological feature vector corresponding to other users. Then, the second weighting parameter corresponding to the target user is used to perform weighted fusion of the first meteorological feature vector of the target user and the second meteorological feature vector of at least one other user to obtain the target meteorological feature vector corresponding to the target user.

[0109] Step 502: Input the target meteorological feature vector and historical electricity consumption into the preset electricity consumption prediction model to determine the predicted electricity consumption of the target user.

[0110] Among them, the preset electricity consumption prediction model can be obtained by iteratively training based on the sample training data of multiple users to predict electricity consumption. It is a general electricity consumption prediction model, such as the Long Short-Term Memory (LSTM) network model.

[0111] Continue to refer to Figure 4 As shown, when forecasting electricity consumption, the target meteorological feature vector corresponding to the target user and the historical electricity consumption corresponding to the target user can be input into a general electricity consumption forecasting model to forecast electricity consumption, learn the correlation between electricity consumption habits and weather, output high-precision electricity consumption forecast results, and finally output the predicted electricity consumption of the target user.

[0112] In this embodiment, for different users, multi-source meteorological data can be fused based on the meteorological fusion parameters of different users. Based on the target meteorological feature vector obtained after fusion, combined with the user's historical electricity consumption, the user's electricity consumption can be predicted. Since the fused meteorological data can better reflect the user's electricity consumption habits and electricity demand, the accuracy of the prediction of the user's electricity consumption can be improved.

[0113] To verify the above method, we used electricity users of a certain electricity sales company as an example. In the data preparation stage, we collected electricity consumption data for a certain year, including meteorological data such as temperature, humidity, and radiation from three predictive data sources and two measured data sources during the same period. We completed the data processing and model training process in the above method, obtained the meteorological fusion parameters corresponding to the target meteorological fusion model for each electricity user, and inferred new fused weather features based on the meteorological fusion parameters of the target electricity users to obtain the target meteorological feature vector of the target electricity users. We then used the electricity consumption prediction model to predict electricity consumption to verify the accuracy.

[0114] Verification shows that the error of the fused weather features represented by the target meteorological feature vector can be reduced by 10% to 20% compared with single-source prediction / measured data (especially under extreme weather conditions). The quality of fused weather data is greatly improved, the electricity consumption prediction error is reduced, and this method can generate meteorological prediction data that better matches the electricity consumption patterns of each user, thus reducing the electricity consumption prediction error.

[0115] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0116] Based on the same inventive concept, this application also provides a meteorological data fusion apparatus for implementing the meteorological data fusion method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more meteorological data fusion apparatus embodiments provided below can be found in the limitations of the meteorological data fusion method described above, and will not be repeated here.

[0117] In one exemplary embodiment, such as Figure 6 As shown, a meteorological data fusion device is provided, including: a first acquisition module 601, a first fusion module 602, and a second fusion module 603, wherein:

[0118] The first acquisition module 601 is used to acquire measured meteorological data and predicted meteorological data from multiple meteorological data sources at the current moment.

[0119] The first fusion module 602 is used to perform weighted fusion of measured meteorological data and predicted meteorological data from multiple meteorological data sources according to the first weighting parameter corresponding to the target user, so as to obtain the first meteorological feature vector.

[0120] The second fusion module 603 is used to perform weighted fusion of the first meteorological feature vector and the second meteorological feature vector of at least one other user based on the second weighting parameters of the target user and at least one other user, to obtain the target meteorological feature vector corresponding to the target user; wherein, the second meteorological feature vector is obtained by weighted fusion of measured meteorological data and predicted meteorological data from multiple meteorological data sources based on the weighting parameters corresponding to other users.

[0121] In one embodiment, the first weighting parameter includes a data weighting parameter and a data source weighting parameter, and the first fusion module 602 includes:

[0122] The first fusion unit is used to perform weighted fusion of measured meteorological data and predicted meteorological data from various meteorological data sources according to data weighting parameters, so as to obtain the third meteorological feature vector of the meteorological data source.

[0123] The second fusion unit is used to perform weighted fusion of the third meteorological feature vectors of each meteorological data source according to the weighting parameters of the data source, so as to obtain the first meteorological feature vector.

[0124] In one embodiment, the device further includes:

[0125] The second acquisition module is used to acquire sample meteorological data from multiple users; the sample meteorological data includes measured sample data and predicted sample data from multiple meteorological data sources at multiple times.

[0126] The feature extraction module is used to input the user's sample meteorological data into a preset feature extraction model for each user, and determine the measured feature vector and predicted feature vector of each meteorological data source of the user at each time.

[0127] The feature fusion module is used to input the measured feature vectors and predicted feature vectors of each meteorological data source of each user at each time into the initial meteorological fusion network to obtain the sample fusion feature vectors at each time. The initial meteorological fusion network is a fusion network based on a hybrid attention mechanism.

[0128] The training module is used to iteratively train the initial meteorological fusion network based on the sample fusion feature vectors at each time point, the historical electricity consumption of the target user at each time point, and the preset electricity consumption prediction model, so as to obtain the target meteorological fusion model corresponding to the target user; the target meteorological fusion model includes the first weighting parameter and the second weighting parameter.

[0129] In one embodiment, the feature extraction module is specifically used to determine the basic data features of the sample meteorological data; the basic data features include at least one of the meteorological value at the current moment, the average meteorological value over a preset time period before the current moment, and the meteorological rate of change; the sample meteorological data and the basic data features are input into a preset feature extraction model to determine the measured feature vector and predicted feature vector of each meteorological data source of the user at each moment.

[0130] In one embodiment, the initial meteorological fusion network includes a first fusion network, a second fusion network, and a third fusion network. The feature fusion module is specifically used to input the measured feature vector and the predicted feature vector at each time moment into the first fusion network for each meteorological data source of each user, to obtain the first fusion feature vector at each time moment; input the first fusion feature vector of each meteorological data source of the user into the second fusion network for each time moment, to obtain the second fusion feature vector at each time moment; and input the second fusion feature vector of each user into the third fusion network to obtain the sample fusion feature vector at each time moment.

[0131] In one embodiment, the training module is specifically used to determine the historical electricity consumption sequence corresponding to each time based on the historical electricity consumption of the target user at each time; input the sample fusion feature vector of each time and the historical electricity consumption sequence of each time into the preset electricity consumption prediction model to obtain the predicted electricity consumption at each time; and iteratively train the initial meteorological fusion network based on the error between the predicted electricity consumption at each time and the historical electricity consumption at each time to obtain the target meteorological fusion model corresponding to the target user.

[0132] In one embodiment, the second acquisition module is specifically used to acquire initial meteorological data from multiple users; the initial meteorological data includes initial measured data and initial forecast data from multiple meteorological data sources at multiple times; the initial meteorological data is preprocessed to obtain sample meteorological data for each user; the preprocessing includes at least one of time alignment processing, spatial alignment processing, and standardization processing.

[0133] In one embodiment, the device further includes:

[0134] The third acquisition module is used to acquire the target user's historical electricity consumption up to the current moment;

[0135] The electricity consumption forecasting module is used to input the target meteorological feature vector and historical electricity consumption into the preset electricity consumption forecasting model to determine the predicted electricity consumption of the target user.

[0136] Each module in the aforementioned meteorological data fusion device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0137] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 7 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores meteorological data from various meteorological data sources and the determined meteorological fusion parameters for each user. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a meteorological data fusion method.

[0138] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0139] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the meteorological data fusion method in any of the above embodiments.

[0140] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the meteorological data fusion method in any of the above embodiments.

[0141] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the meteorological data fusion method in any of the above embodiments.

[0142] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0143] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0144] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0145] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A meteorological data fusion method, characterized in that, The method includes: Acquire measured and predicted meteorological data from multiple meteorological data sources at the current moment; Based on the first weighting parameter corresponding to the target user, the measured meteorological data and predicted meteorological data from the multiple meteorological data sources are weighted and fused to obtain the first meteorological feature vector; Based on the second weighting parameter between the target user and at least one other user, the first meteorological feature vector and the second meteorological feature vector of the at least one other user are weighted and fused to obtain the target meteorological feature vector corresponding to the target user; wherein, the second meteorological feature vector is obtained by weighting and fusing measured meteorological data and predicted meteorological data from multiple meteorological data sources based on the weighting parameter corresponding to the other user; the other user is at least one user with a strong correlation with the target user, and the correlation is considered from at least one dimension of electricity consumption similarity, user feature similarity, user relationship, user type, and regional location; The method further includes: Acquire sample meteorological data from multiple users; the sample meteorological data includes measured sample data and predicted sample data from multiple meteorological data sources at multiple times; For each user, the sample meteorological data of the user is input into a preset feature extraction model to determine the measured feature vector and predicted feature vector of each meteorological data source of the user at each time. The measured feature vectors and predicted feature vectors of the meteorological data sources of each user at each time are input into the initial meteorological fusion network to obtain the sample fusion feature vectors at each time; the initial meteorological fusion network is a fusion network based on a hybrid attention mechanism; Based on the sample fusion feature vectors at each time point, the historical electricity consumption of the target user at each time point, and the preset electricity consumption prediction model, the initial meteorological fusion network is iteratively trained to obtain the target meteorological fusion model corresponding to the target user; the target meteorological fusion model includes the first weighting parameter and the second weighting parameter.

2. The method according to claim 1, characterized in that, The first weighting parameter includes a data weighting parameter and a data source weighting parameter. The step of weighting and fusing the measured meteorological data and predicted meteorological data from the multiple meteorological data sources according to the first weighting parameter corresponding to the target user to obtain a first meteorological feature vector includes: For each of the meteorological data sources, the measured meteorological data and predicted meteorological data of the meteorological data sources are weighted and fused according to the data weighting parameters to obtain the third meteorological feature vector of the meteorological data source. Based on the weighting parameters of the data sources, the third meteorological feature vectors of each meteorological data source are weighted and fused to obtain the first meteorological feature vector.

3. The method according to claim 1, characterized in that, The step of inputting the user's sample meteorological data into a preset feature extraction model to determine the measured feature vector and predicted feature vector of each of the user's meteorological data sources at each time point includes: Determine the basic characteristics of the sample meteorological data; the basic characteristics include at least one of the meteorological value at the current moment, the average meteorological value over a preset time period prior to the current moment, and the meteorological rate of change; The sample meteorological data and the basic features of the data are input into a preset feature extraction model to determine the measured feature vector and predicted feature vector of each meteorological data source of the user at each time.

4. The method according to claim 1, characterized in that, The initial meteorological fusion network includes a first fusion network, a second fusion network, and a third fusion network. The step of inputting the measured feature vectors and predicted feature vectors of each meteorological data source from each user at each time moment into the initial meteorological fusion network to obtain the sample fusion feature vectors at each time moment includes: For each of the meteorological data sources of each user, the measured feature vector and the predicted feature vector at each time are input into the first fusion network to obtain the first fusion feature vector at each time. For each of the aforementioned times, the first fusion feature vector of each of the aforementioned meteorological data sources of the user is input into the second fusion network to obtain the second fusion feature vector of each of the aforementioned times; The second fusion feature vector of each user is input into the third fusion network to obtain the sample fusion feature vector at each time point.

5. The method according to claim 1, characterized in that, The initial meteorological fusion network is iteratively trained based on the sample fusion feature vectors at each of the stated times, the historical electricity consumption of the target user at each of the stated times, and a preset electricity consumption prediction model to obtain the target meteorological fusion model corresponding to the target user, including: Based on the target user’s historical electricity consumption at each of the said times, determine the historical electricity consumption sequence corresponding to each of the said times; The sample fusion feature vectors and historical electricity consumption sequences at each of the aforementioned times are input into a preset electricity consumption prediction model to obtain the predicted electricity consumption at each of the aforementioned times. Based on the error between the predicted electricity consumption at each time point and the historical electricity consumption at each time point, the initial meteorological fusion network is iteratively trained to obtain the target meteorological fusion model corresponding to the target user.

6. The method according to claim 1, characterized in that, The acquisition of sample meteorological data from multiple users includes: Acquire initial meteorological data from multiple users; the initial meteorological data includes initial measured data and initial forecast data from multiple meteorological data sources at multiple times; The initial meteorological data is preprocessed to obtain sample meteorological data for each user; the preprocessing includes at least one of time alignment processing, spatial alignment processing, and standardization processing.

7. The method according to any one of claims 1 to 6, characterized in that, The method further includes: Obtain the target user's historical electricity consumption up to the current moment; The target meteorological feature vector and the historical electricity consumption are input into a preset electricity consumption prediction model to determine the predicted electricity consumption of the target user.

8. A meteorological data fusion device, characterized in that, The device includes: The first acquisition module is used to acquire measured meteorological data and forecast meteorological data from multiple meteorological data sources at the current moment; The first fusion module is used to perform weighted fusion of measured meteorological data and predicted meteorological data from the multiple meteorological data sources according to the first weighting parameter corresponding to the target user, so as to obtain a first meteorological feature vector. The second fusion module is used to perform weighted fusion of the first meteorological feature vector and the second meteorological feature vector of the at least one other user based on the second weighting parameters of the target user and at least one other user, to obtain the target meteorological feature vector corresponding to the target user; wherein, the second meteorological feature vector is obtained by weighted fusion of measured meteorological data and predicted meteorological data from multiple meteorological data sources based on the weighting parameters corresponding to the other user; the other user is at least one user with a strong correlation with the target user, and the correlation is considered from at least one dimension of electricity consumption similarity, user feature similarity, user relationship, user type, and regional location; The second acquisition module is used to acquire sample meteorological data from multiple users; the sample meteorological data includes measured sample data and predicted sample data from multiple meteorological data sources at multiple times. The feature extraction module is used to input the sample meteorological data of each user into a preset feature extraction model for each user, and determine the measured feature vector and predicted feature vector of each meteorological data source of the user at each time. The feature fusion module is used to input the measured feature vectors and predicted feature vectors of the meteorological data sources of each user at each time into the initial meteorological fusion network to obtain the sample fusion feature vectors at each time; the initial meteorological fusion network is a fusion network based on a hybrid attention mechanism; The training module is used to iteratively train the initial meteorological fusion network based on the sample fusion feature vectors at each time point, the historical electricity consumption of the target user at each time point, and a preset electricity consumption prediction model, to obtain the target meteorological fusion model corresponding to the target user; the target meteorological fusion model includes the first weighting parameter and the second weighting parameter.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images