A power load prediction method, device, equipment and product

Abstract

This invention discloses a method, apparatus, device, and product for power load forecasting. The method includes: acquiring a dataset; determining clustering results based on the dataset using a scenario segmentation submodule in a power load forecasting model, wherein the clustering results indicate the mapping relationship between features of various set weather events and corresponding load change patterns; correcting the dataset for temperature using a temperature accumulation effect correction submodule in the power load forecasting model to determine the corrected effective temperature; and performing power load forecasting based on the dataset, the effective temperature, and the clustering results using an adaptive decomposition forecasting submodule in the power load forecasting model, outputting the load forecasting result. The technical solution of this invention addresses the problem of insufficient ability of power load forecasting models to represent temperature accumulation effects, improves the forecasting accuracy of power load data, and provides accurate and reliable technical support for power grid dispatching and risk prevention.

Description

Technical Field

[0001] This invention relates to the field of power load forecasting technology, and in particular to a power load forecasting method, apparatus, equipment and product. Background Technology

[0002] With the intensification of global climate change and the advancement of the construction of new power systems, the impact of extreme weather events on power load is becoming increasingly prominent. Load curves exhibit complex characteristics such as violent fluctuations, strong non-stationarity, and peak variation, posing severe challenges to the real-time balance and safe regulation of the power grid.

[0003] Traditional methods generally rely on a single model structure and lack the ability to correct for temperature under continuous high-temperature weather. Furthermore, under such weather conditions, the load sequence exhibits strong non-stationarity and high-frequency abrupt changes, resulting in insufficient ability of the model to capture peak loads and significant prediction biases. Consequently, they cannot provide accurate and reliable technical support for power grid dispatching and risk prevention. Summary of the Invention

[0004] This invention provides a method, apparatus, equipment, and product for predicting power load, in order to solve the problem of insufficient accuracy in predicting power load under extreme weather conditions.

[0005] According to one aspect of the present invention, a power load forecasting method is provided, comprising: Obtain a dataset, which includes historical power load data and meteorological data under a set weather event, wherein the set weather event includes weather events that have an impact on power load; The scenario segmentation submodule in the power load forecasting model determines the clustering results based on the dataset. The clustering results indicate the mapping relationship between the characteristics of various set weather events and the corresponding load change patterns. The effective temperature after correction is determined by using the temperature accumulation effect correction submodule in the power load prediction model to correct the temperature of the dataset. The adaptive decomposition prediction submodule in the power load prediction model performs power load prediction based on the dataset, the effective temperature, and the clustering results, and outputs the load prediction results, which indicate the predicted power load under the set weather event.

[0006] According to another aspect of the present invention, an electricity load forecasting device is provided, comprising: The acquisition module is used to acquire a dataset, which includes historical power load data and meteorological data under a set weather event, and the set weather event includes weather events that have an impact on the power load; The clustering result determination module is used to determine the clustering result based on the dataset through the scenario segmentation submodule in the power load prediction model. The clustering result indicates the mapping relationship between the characteristics of various set weather events and the corresponding load change patterns. The temperature correction module is used to correct the temperature of the dataset through the temperature accumulation effect correction submodule in the power load prediction model, and to determine the corrected effective temperature. The output module is used to perform power load prediction based on the dataset, the effective temperature and the clustering results through the adaptive decomposition prediction submodule in the power load prediction model, and output the load prediction result, which indicates the prediction result of the power load under the set weather event.

[0007] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the method described in any embodiment of the present invention.

[0008] According to another aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program that, when executed by a processor, implements the method described in any embodiment of the present invention.

[0009] The technical solution of this invention involves acquiring a dataset, which includes historical power load data and meteorological data under a set weather event, wherein the set weather event includes weather events that affect power load; determining clustering results based on the dataset using a scenario segmentation submodule in the power load prediction model, wherein the clustering results indicate the mapping relationship between the characteristics of various set weather events and corresponding load change patterns; correcting the temperature of the dataset using a temperature accumulation effect correction submodule in the power load prediction model to determine the corrected effective temperature; and performing power load prediction using an adaptive decomposition prediction submodule in the power load prediction model based on the dataset, the effective temperature, and the clustering results, outputting a load prediction result, wherein the load prediction result indicates the predicted power load under the set weather event. The technical solution of this invention determines the clustering results through a scenario segmentation submodule. The clustering results indicate the mapping relationship between the characteristics of a set weather event and the corresponding load change pattern, improving the scenario segmentation accuracy of the adaptive decomposition prediction submodule, laying the foundation for data-driven modeling, and improving the ability to characterize extreme load peaks and fluctuations under set weather events. A temperature accumulation effect correction submodule performs temperature correction to determine the corrected effective temperature, solving the problem of insufficient characterization of temperature accumulation effects in power load prediction models and improving the prediction accuracy of power load data. Finally, the adaptive decomposition prediction submodule outputs load prediction results, further improving the prediction accuracy of power load data under set weather events.

[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of a power load forecasting method provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of a method for determining clustering results provided in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the prediction process of an adaptive decomposition prediction submodule provided in an embodiment of the present invention. Figure 4This is a flowchart of a power load forecasting method provided in Embodiment 2 of the present invention; Figure 5 This is a flowchart of a temperature correction method provided in an embodiment of the present invention; Figure 6 This is a flowchart of a power load forecasting method provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of a power load forecasting device provided in Embodiment 3 of the present invention; Figure 8 A schematic diagram of an electronic device that can be used to implement embodiments of the present invention is shown. Detailed Implementation

[0013] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0014] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0015] Example 1 Figure 1 This is a flowchart of a power load forecasting method provided in Embodiment 1 of the present invention. This embodiment is applicable to situations where power load forecasting is performed under specified weather events. The method can be executed by a power load forecasting device, which can be implemented in hardware and / or software and can be configured in an electronic device. For example, the electronic device can be a computer or a server. Figure 1 As shown, the method includes: S110. Obtain a dataset, which includes historical power load data and meteorological data under a set weather event, wherein the set weather event includes weather events that have an impact on the power load.

[0016] In this embodiment, the dataset can be a dataset including historical power load data under a set weather event and corresponding meteorological data. The set weather event can be a weather event that causes power load fluctuations to exceed a preset range. The set weather event can be continuous high temperatures, heavy rainfall, blizzards, or frost, etc. Continuous high temperatures can be a set weather event where the temperature is higher than a preset temperature and remains there for a certain number of days. Historical power load data can be time-series data used to describe power load conditions. Historical power load data includes, but is not limited to, multiple timestamps and the power consumption corresponding to each timestamp. Historical power load data can fluctuate beyond a preset range due to the set weather event; therefore, improving the prediction accuracy of power load under the set weather event can effectively provide technical support for power grid dispatching and risk prevention. Meteorological data can be meteorological data corresponding to the set weather event. Meteorological data can be time-series data. Meteorological data includes, but is not limited to, temperature, precipitation, wind speed, air pressure, and humidity. The source of meteorological data can be, for example, ground weather stations and / or radar.

[0017] Specifically, historical power load data and meteorological data collected under specified weather events are added to the dataset. The historical power load data and meteorological data are aligned according to time sequence indicated by timestamps. The aligned dataset is then denoised using methods such as wavelet transform or sliding filter. The denoised dataset is then normalized to map the data to a uniform numerical range, ensuring consistent data length and eliminating the influence of units. Furthermore, features of the data can be extracted and added to the dataset.

[0018] S120. The scenario segmentation submodule in the power load prediction model determines the clustering results based on the dataset. The clustering results indicate the mapping relationship between the characteristics of various set weather events and the corresponding load change patterns.

[0019] In this embodiment, the power load forecasting model can be a model used to predict power load over a period of time based on a dataset. The power load forecasting model can classify set weather events, correct for temperature accumulation effects, and predict power load over a future period based on the classification of set weather events and the corrected temperature. The power load forecasting model can include a scenario segmentation submodule. The scenario segmentation submodule can be a module used to identify and model the correlation between set weather events and power load change patterns. The scenario segmentation submodule can output clustering results based on the dataset. The clustering results can be the result obtained after clustering the data in the dataset. The clustering results can include scenario category labels, corresponding clusters, and typical load curve cluster centers. Scenario category labels can be labels indicating the set weather event to which a cluster belongs. Typical load curve cluster centers can be representative change curves in shape representing all historical power load data in the cluster. Typical load curve cluster centers can be the baseline curve corresponding to the scenario category label. The mapping relationship can describe the relationship between set weather events and power load changes. The mapping relationship can be expressed as a mapping relationship between the characteristics of set weather events and the power load change curves. The load change pattern can be the power load change pattern. Load variation patterns can be the fluctuations in power load over time. Load variation patterns can also be the patterns of change in power load caused by predetermined weather events.

[0020] Specifically, the scene segmentation submodule clusters the dataset. During the clustering process, the number of multiple clusters is first determined. Based on the average distance between samples in each cluster and samples in other clusters, as well as the average distance between samples in the same cluster, the optimal number of clusters is determined. The clustering results are determined and output based on the optimal number of clusters.

[0021] S130. The dataset is corrected for temperature using the temperature accumulation effect correction submodule in the power load prediction model, and the corrected effective temperature is determined.

[0022] In this embodiment, the temperature accumulation effect correction submodule can be a module that corrects the temperature in the dataset. When a weather event indicates continuous high temperatures, traditional correction strategies are difficult to accurately reflect the lag relationship between perceived temperature and load growth; therefore, the temperature accumulation effect correction submodule is needed to perform temperature correction. The effective temperature can be the corrected temperature. The effective temperature can be the sum of the temperature change and the corresponding uncorrected temperature in the dataset.

[0023] Specifically, the temperature accumulation effect correction submodule extracts the time-series temperature data from the samples in the input dataset, performs fuzzy inference on the time-series temperature data to determine the amount of temperature change, and determines the corrected effective temperature by summing the amount of temperature change with the corresponding temperature before correction.

[0024] S140. The adaptive decomposition prediction submodule in the power load prediction model performs power load prediction based on the dataset, the effective temperature, and the clustering results, and outputs the load prediction results, which indicate the prediction results of the power load under the set weather event.

[0025] In one embodiment, the adaptive decomposition prediction submodule can be a module for predicting power load. The adaptive decomposition prediction submodule can predict power load based on the dataset, effective temperature, and clustering results.

[0026] Specifically, the adaptive decomposition prediction submodule determines the set weather event corresponding to the data in the dataset based on the clustering results. If the set weather event includes continuous high temperature, the effective temperature is used as one of the input features and normalized together with the dataset. The normalized data is decomposed into multiple components, and power load prediction is performed based on the different components. The prediction results of each component are fused together to output the load prediction result.

[0027] The technical solution of this invention involves acquiring a dataset, which includes historical power load data and meteorological data under a set weather event, wherein the set weather event includes weather events that affect power load; determining clustering results based on the dataset using a scenario segmentation submodule in the power load prediction model, wherein the clustering results indicate the mapping relationship between the characteristics of various set weather events and corresponding load change patterns; correcting the temperature of the dataset using a temperature accumulation effect correction submodule in the power load prediction model to determine the corrected effective temperature; and performing power load prediction using an adaptive decomposition prediction submodule in the power load prediction model based on the dataset, the effective temperature, and the clustering results, outputting a load prediction result, wherein the load prediction result indicates the predicted power load under the set weather event. The technical solution of this invention determines the clustering results through a scenario segmentation submodule. The clustering results indicate the mapping relationship between the characteristics of a set weather event and the corresponding load change pattern, improving the scenario segmentation accuracy of the adaptive decomposition prediction submodule, laying the foundation for data-driven modeling, and improving the ability to characterize extreme load peaks and fluctuations under set weather events. A temperature accumulation effect correction submodule performs temperature correction to determine the corrected effective temperature, solving the problem of insufficient characterization of temperature accumulation effects in power load prediction models and improving the prediction accuracy of power load data. Finally, the adaptive decomposition prediction submodule outputs load prediction results, further improving the prediction accuracy of power load data under set weather events.

[0028] In one embodiment, the step of determining the clustering result based on the dataset using the scenario segmentation submodule in the power load prediction model includes: S1201. The scene segmentation submodule categorizes the data in the dataset into multiple categories based on a preset distance threshold range, and determines the search range based on the categories.

[0029] In this embodiment, the distance threshold interval can be a threshold interval of the distance between a sample in the dataset and the center sample. The distance threshold interval can be determined empirically, and this invention does not impose any limitations on it. The distance can be, for example, Euclidean distance. The center sample can be any sample in the dataset. The category can be the category of the data in the dataset. The category can be determined based on the distance threshold interval. The search interval can be a set of integers determined based on the categories. The search interval can be directly determined by the number of categories. For example, when the number of categories is m, the search interval can be set to [m, m+n], where n can be an extended value set based on experience.

[0030] Specifically, any sample is selected as the center sample, and the distance between other samples in the dataset and the center sample is determined. If the distance is less than the upper limit of the distance threshold interval, the corresponding sample is classified into one class. At the same time, samples whose distance is less than the lower limit of the distance threshold interval are removed from the candidates of the center sample, and finally multiple classes are obtained. The search interval is determined according to the number of classes.

[0031] For example, the distance threshold interval is set to [T2, T1]. Samples are iteratively selected as center samples. Samples with a distance less than T1 from the center sample are classified into one class, and samples with a distance less than T2 are removed from the candidates of the center sample. This process continues until all samples have been processed. The number of classes obtained determines the search interval for subsequent clustering.

[0032] S1202. The scenario segmentation submodule traverses the number of candidate clusters in the search interval. For each number of candidate clusters, the candidate clustering result corresponding to each number of candidate clusters is determined. The first average distance from each sample in the cluster to samples in other clusters in each candidate cluster result is determined, as well as the second average distance from each sample in the cluster to other samples in the cluster.

[0033] In this embodiment, the number of candidate clusters can be an integer within the search interval. The number of candidate clusters indicates the number of clusters into which the samples in the dataset are divided during clustering. There can be multiple candidate clusters. The candidate clustering result can be the clustering result determined based on the number of candidate clusters. The candidate clustering result can be obtained by dividing the samples in the dataset according to a preset number of candidate clusters, forming a corresponding number of clusters. The first average distance can be the average distance from each sample within a certain cluster to samples in other clusters in the candidate clustering result. The first average distance reflects the average separation degree of each sample within a certain cluster from samples in other clusters; the higher the average separation degree, the more obvious the difference between the clusters. The first average distance can also be denoted as the inter-cluster average distance. The second average distance can be the average distance from each sample within a cluster to other samples within the cluster. The second average distance reflects the tightness of clustering within a cluster; the smaller the second average distance, the tighter the clustering of samples within the cluster. The second average distance can also be denoted as the intra-cluster average distance.

[0034] Specifically, the number of candidate clusters in the search interval is traversed, and the corresponding candidate clustering result is determined based on the number of candidate clusters. The first average distance and the second average distance of each sample in each candidate clustering result are calculated.

[0035] S1203. The scene segmentation submodule determines the contour coefficient of each sample in each of the candidate clustering results based on the first average distance and the second average distance.

[0036] In this embodiment, the silhouette coefficient can be an evaluation metric used to quantify each sample in the dataset. The silhouette coefficient value can be between 1 and -1.

[0037] Specifically, the silhouette coefficients of each sample in each candidate clustering result are determined based on the first average distance and the second average distance.

[0038] For example, the profile coefficient can be determined by the following formula: in, It can be the first in the dataset The silhouette coefficients corresponding to each sample It could be the first average distance. It can be the second average distance.

[0039] S1204. The average value of the silhouette coefficient of each sample in each candidate clustering result is determined by the scene segmentation submodule. The candidate cluster number corresponding to the largest average value among the average values ​​is taken as the final cluster number, and the clustering result corresponding to the final cluster number is determined.

[0040] In this embodiment, the final number of clusters can be a finalized number that optimizes the average silhouette coefficient. The final number of clusters can be determined based on the silhouette coefficient of each sample in each candidate clustering result.

[0041] Specifically, the average or sum of the silhouette coefficients of each sample in each candidate clustering result is determined, the number of candidate clusters corresponding to the largest average or sum is determined as the final number of clusters, and the corresponding clustering result is determined based on the final number of clusters.

[0042] For example, Figure 2 This is a flowchart of a method for determining clustering results provided by an embodiment of the present invention. First, the scene segmentation submodule groups the data in the dataset into multiple categories according to a preset distance threshold interval. A search interval is determined based on the categories, and the value range of the search interval includes [kmin, kma], where k is the number of candidate clusters. Second, the number of candidate clusters is traversed, starting from the minimum value kma in the search interval, to determine the corresponding candidate clustering results and the silhouette coefficient of each sample in each candidate clustering result, until the maximum value kma in the search interval is reached. The average or sum of the silhouette coefficients of each sample corresponding to each number of candidate clusters is used to determine the clustering result corresponding to the final number of clusters.

[0043] In one embodiment, the first average distance includes the average shape distance of each sample within a cluster to samples in other clusters in the clustering result; the second average distance includes the average shape distance of each sample within a cluster to other samples in the cluster group.

[0044] In this embodiment, shape distance can be a distance used to describe the morphological differences between any two samples in the dataset. Since the samples in the dataset are time-series data, shape distance can effectively capture the fluctuations in power load under set weather events, improve the sensitivity of the power load prediction model to the fluctuations in power load under set weather events, and achieve refined classification of power load according to set weather events.

[0045] Specifically, the first and second average distances can be determined using shape distance. For example, the cross-correlation coefficient between two samples can be determined first, and the difference between 1 and the correlation coefficient can be used as the shape distance.

[0046] For example, for two samples in the dataset and ,in and All of these are time-series data. The cross-correlation coefficient can be expressed as follows: in, It can be the cross-correlation coefficient. It could be displacement. It can be and In displacement The interrelation, For sequence and Zero displacement autocorrelation.

[0047] In one embodiment, the adaptive decomposition prediction submodule in the power load prediction model performs power load prediction based on the dataset, the effective temperature, and the clustering results, and outputs load prediction results, including: S1401. The adaptive decomposition prediction submodule determines, based on the clustering results, whether the set weather events to which the samples in the dataset belong include continuous high-temperature events, wherein the continuous high-temperature events include events in which the temperature is higher than a set threshold for a continuous set duration; if so, the dataset and the effective temperature are used as input features, and the input features are decomposed into multiple intrinsic mode components and residuals.

[0048] In this embodiment, a continuous high-temperature event can be a set weather event in which the daily maximum temperature is consistently higher than a pre-set threshold for a continuous preset duration. In power load forecasting, continuous high-temperature events cause a temperature accumulation effect, which leads to the daily accumulation of power load, making the curve shape of the power load significantly different from that of short-term single-day high temperatures. The set duration can be the continuous duration of the dates used to determine whether a continuous high-temperature event has occurred. The set threshold can be the temperature threshold used to determine whether a continuous high-temperature event has occurred. The set duration and set threshold can be determined empirically, and this invention does not impose any limitations on them. The input features can be the features input to the adaptive decomposition prediction submodule. When the set weather event to which the samples in the dataset belong includes a continuous high-temperature event, the input features can include the effective temperature. The intrinsic mode components can be multiple components separated from the input features. The intrinsic mode components can have different occurrence frequencies. The noise of intrinsic mode components with an occurrence frequency higher than a certain threshold has a significant impact on power load forecasting, so noise reduction processing is required.

[0049] Specifically, the system determines whether the samples in the dataset belong to a given weather event that includes consecutive high-temperature events. If so, the dataset and the effective temperature are used as input features. The adaptive decomposition prediction submodule normalizes the input features before inputting them into the model within the adaptive decomposition prediction submodule. The model in the adaptive decomposition prediction submodule decomposes the input features into multiple intrinsic mode components and residuals.

[0050] For example, after normalizing the input sequence, the input sequence is constructed into input-output samples using a sliding window. In the input model, the input sequence is denoted as... After decomposition, the intrinsic mode components are obtained. and residual ,in For the first Each intrinsic mode component.

[0051] S1402. The sample entropy of the intrinsic mode component and the residual is determined by the adaptive decomposition prediction submodule. Based on the numerical relationship between each sample entropy and the sample entropy threshold, the intrinsic mode component and the residual corresponding to each sample entropy are divided into high-frequency components and low-frequency components.

[0052] In this embodiment, sample entropy can be entropy used to assess the frequency of occurrence. Based on sample entropy, intrinsic mode components and residuals can be divided into high-frequency components and low-frequency components. Sample entropy threshold can be a threshold used to distinguish between high-frequency and low-frequency components. The sample entropy threshold can be expressed as the product of an empirical coefficient and the average of all sample entropies. High-frequency components can be intrinsic mode components whose sample entropy is not less than the sample entropy threshold. High-frequency components contain noise and other interference. Low-frequency components can be intrinsic mode components and / or residuals whose sample entropy is less than the sample entropy threshold. Low-frequency components have less interference and are easier to predict than high-frequency components.

[0053] Specifically, the sample entropy of the intrinsic mode components and residuals is determined. Based on the relationship between the sample entropy and the sample entropy threshold, the intrinsic mode components and residuals are divided into high-frequency components or low-frequency components. For high-frequency components, additional noise reduction processing is required.

[0054] For example, the intrinsic mode components and residuals can be separated based on the sample entropy using a discriminant function, as shown in the following equation: in, Let be the discriminant function for the i-th intrinsic mode component. Let be the sample entropy of the i-th intrinsic mode component. This is an empirical parameter and can be set to 0.5. When When the value equals 1, the corresponding intrinsic mode components are high-frequency components. When the value equals 0, the corresponding intrinsic mode components are low-frequency components. For the residuals, they can be directly divided into low-frequency components. For high-frequency components, the variational mode decomposition (VMD) method can be used to first subdivide and then synthesize them to obtain the denoised high-frequency sequence.

[0055] S1403. The adaptive decomposition prediction submodule determines the corresponding high-frequency component prediction result and low-frequency component prediction result based on the high-frequency component and the low-frequency component, respectively, and fuses the high-frequency component prediction result and the low-frequency component prediction result to obtain the load prediction result.

[0056] In this embodiment, the high-frequency component prediction result can be the prediction result determined by the adaptive decomposition prediction submodule based on the high-frequency component. The high-frequency component prediction result can reflect the predicted value of the rapidly changing part of the power load. The low-frequency component prediction result can be the prediction result determined by the adaptive decomposition prediction submodule based on the low-frequency component. The low-frequency component prediction result can reflect the predicted value of the stable or regularly changing part of the power load. The high-frequency component prediction result and the low-frequency component prediction result are complementary in the time domain.

[0057] Specifically, the adaptive decomposition prediction submodule includes two independent model structures, such as branches, for parallel processing of the input high-frequency and low-frequency components, and outputting the corresponding high-frequency and low-frequency component prediction results respectively. The output of this model structure can be connected to the input of a self-attention model, which can then fuse the high-frequency and low-frequency component prediction results to obtain the load prediction result.

[0058] For example, the high-frequency component prediction results and the low-frequency component prediction results are mapped to a query matrix. Key matrix AND-value matrix The high-frequency component prediction results and the low-frequency component prediction results are fused using a self-attention method, as shown in the following equation: in, Using the key vector dimension, softmax normalizes the relevance weights in the time dimension, highlighting the key moments that trigger weather events.

[0059] In one embodiment, determining the corresponding high-frequency component prediction result and low-frequency component prediction result based on the high-frequency component and low-frequency component, respectively, includes: A1. Denoise the high-frequency components to obtain the denoised high-frequency components.

[0060] In this embodiment, the high-frequency component after noise reduction can be a high-frequency component that has undergone noise reduction processing. The noise reduction processing method can be to first divide the high-frequency component into multiple components and then merge them.

[0061] Specifically, the high-frequency components are divided into multiple components. After filtering out the components that include noise, the remaining components are then merged to obtain the noise-reduced high-frequency components.

[0062] A2. The noise-reduced high-frequency component and the low-frequency component are respectively spliced ​​with the corresponding meteorological features to obtain spliced ​​high-frequency components and spliced ​​low-frequency components. The meteorological features include meteorological features extracted from the meteorological data.

[0063] In this embodiment, the meteorological features can be features extracted from meteorological data. Meteorological features can characterize the meteorological changes caused by a given weather event. The stitched high-frequency components can be a data matrix including the denoised high-frequency components and meteorological features. The stitched high-frequency components can be the result of stitching the denoised high-frequency components and meteorological features along the feature dimension. The stitched low-frequency components can be a data matrix including low-frequency components and meteorological features. The stitched low-frequency components can be the result of fusing and connecting the denoised high-frequency components and meteorological features along the feature dimension.

[0064] Specifically, the noise-reduced high-frequency and low-frequency components are spliced ​​with the corresponding meteorological features along the feature dimension to obtain the spliced ​​high-frequency and low-frequency components.

[0065] A3. Input the spliced ​​high-frequency component and the spliced ​​low-frequency component into different branches of the dual-branch prediction model, and each branch outputs the prediction results of the high-frequency component and the low-frequency component, respectively.

[0066] In this embodiment, the dual-branch prediction model can be a neural network model comprising two parallel structures. The dual-branch prediction model can include two structurally independent branches that do not share parameters, used to process the spliced ​​high-frequency components and the spliced ​​low-frequency components respectively.

[0067] Specifically, the spliced ​​high-frequency components and spliced ​​low-frequency components are input into different branches of the dual-branch prediction model, and each branch performs prediction in parallel to obtain the corresponding high-frequency component prediction results and low-frequency component prediction results.

[0068] For example, Figure 3 This is a flowchart illustrating the prediction process of an adaptive decomposition prediction submodule provided in this embodiment of the invention. First, input features are constructed. Then, the input features are decomposed into multiple intrinsic mode components (IMFs) and residuals. These IMFs and residuals are further divided into high-frequency components and low-frequency components. The high-frequency components are denoised to obtain denoised high-frequency components. The denoised high-frequency and low-frequency components are then input into a dual-branch prediction model. The dual-branch prediction model outputs prediction results for both high-frequency and low-frequency components. These prediction results are then fused to obtain the load prediction result. The training process of the dual-branch prediction model can be as follows: The output of the untrained dual-branch prediction model on the training set is obtained; the weighted mean absolute error (MAE) and weighted root mean square error (RMSE) loss of the output are calculated; and the trained dual-branch prediction model is output when early stopping or convergence conditions are met.

[0069] Example 2 Figure 4 This is a flowchart of a power load forecasting method provided in Embodiment 2 of the present invention. This embodiment is an optimization based on any of the above embodiments, and mainly includes a detailed description of the process of temperature correction of the dataset. It should be noted that technical details not described in detail in this embodiment can be found in any of the above embodiments. Figure 2 As shown, the method includes: S210. Obtain a dataset, which includes historical power load data and meteorological data under a set weather event, wherein the set weather event includes weather events that have an impact on the power load.

[0070] S220. The scenario segmentation submodule in the power load prediction model determines the clustering results based on the dataset. The clustering results indicate the mapping relationship between the characteristics of various set weather events and the corresponding load change patterns.

[0071] S230. Extract a temperature sequence from the dataset using the temperature accumulation effect correction submodule. The temperature sequence includes multiple raw temperatures.

[0072] In this embodiment, the temperature sequence can be a collection of time-series data reflecting temperature changes extracted from a dataset. The temperature sequence may include the original temperature and its corresponding date, as well as a sequence of the highest temperatures for a certain number of days prior to that date. The original temperature can be a measured value of the temperature in the temperature sequence. The original temperature can be used as a reference value for temperature correction.

[0073] Specifically, the temperature accumulation effect correction submodule extracts temperature sequences from the dataset and performs noise reduction, interpolation, and normalization on the temperature sequences.

[0074] S240. The temperature accumulation effect correction submodule determines the continuous high temperature intensity index of the temperature sequence based on a preset high temperature threshold and the temperature sequence.

[0075] In this embodiment, the high-temperature threshold can be a threshold used to determine whether temperature will affect the power load. The high-temperature threshold can be set empirically, and this invention does not impose any limitations on it. The continuous high-temperature intensity index can be an index used to simultaneously reflect the number of consecutive days of high temperature and the magnitude of exceeding the high-temperature threshold. The continuous high-temperature intensity index can be calculated by accumulating the magnitude of the daily actual temperature exceeding the high-temperature threshold within a certain number of days.

[0076] Specifically, calculate the continuous high-temperature intensity index for the date corresponding to each original temperature in the temperature series.

[0077] For example, continuous high temperature intensity index As shown in the following formula: in, Date corresponding to the original temperature The previous chapter The original temperature of the sky, This is the high temperature threshold.

[0078] S250. The temperature sequence and the continuous high temperature intensity index are fuzzified by the temperature accumulation effect correction submodule to obtain the corresponding membership degree.

[0079] Specifically, the temperature sequence and continuous high temperature intensity index are input into the temperature accumulation effect correction submodule, and the temperature sequence and continuous high temperature intensity index are mapped to a preset subset to obtain the corresponding membership degree.

[0080] S260. The membership degree is input into the fuzzy inference model in the temperature accumulation effect correction submodule through the temperature accumulation effect correction submodule to obtain the fuzzy output result.

[0081] In this embodiment, the fuzzy inference model can be a model that outputs fuzzy results based on membership degrees. The fuzzy inference model can evaluate membership degrees according to preset rules to obtain fuzzy output results. The fuzzy output results can be the output results of the fuzzy inference model. The fuzzy output results can be represented as the distribution function of membership degrees on the universe of discourse, or they can be denoted as the membership function. The membership function can be represented as a continuous function or as discrete sampling points.

[0082] Specifically, the membership degree is input into the fuzzy inference model. The fuzzy inference model evaluates the membership degree according to preset rules, and obtains the evaluation results to determine the fuzzy output result.

[0083] S270. The fuzzy output result is defuzzified by the temperature accumulation effect correction submodule to obtain the temperature correction amount, and the sum of the original temperature and the temperature correction amount is determined as the corrected effective temperature.

[0084] In this embodiment, the temperature correction amount can be a compensation value for the original temperature. The temperature correction amount can be obtained by defuzzifying the fuzzy output result.

[0085] Specifically, the temperature correction amount is determined by performing a weighted integral over the universe of discourse or the domain of definition, or by weighted summation of the sampling points, on the fuzzy output result. The universe of discourse or the domain of definition can be the universe of discourse or the domain of definition of the temperature correction amount.

[0086] For example, defuzzification can be represented by the following formula: The membership function in the fuzzy output is: ,in For temperature correction in the domain The total membership function value on, For date The blurred temperature.

[0087] For example, Figure 5 This is a flowchart of a temperature correction method provided in an embodiment of the present invention. The temperature sequence is input into the temperature accumulation effect correction submodule. First, the temperature sequence is fuzzified to obtain membership degrees. Fuzzy inference is performed on the membership degrees according to preset rules to obtain a fuzzy output result. The fuzzy output result is then defuzzified to obtain the temperature correction amount. The sum of the original temperature and the temperature correction amount is determined as the corrected effective temperature.

[0088] S280. The adaptive decomposition prediction submodule in the power load prediction model performs power load prediction based on the dataset, the effective temperature, and the clustering results, and outputs the load prediction result, which indicates the prediction result of the power load under the set weather event.

[0089] The technical solution of this invention involves: acquiring a dataset including historical power load data and meteorological data under a set weather event, wherein the set weather event includes weather events that affect power load; determining clustering results based on the dataset using a scenario segmentation submodule in a power load prediction model, wherein the clustering results indicate the mapping relationship between the characteristics of various set weather events and corresponding load change patterns; extracting temperature sequences from the dataset using a temperature accumulation effect correction submodule; determining a continuous high-temperature intensity index of the temperature sequence based on a preset high-temperature threshold and the temperature sequence; performing fuzzification processing on the temperature sequence and the continuous high-temperature intensity index to obtain corresponding membership degrees; inputting the membership degrees into a fuzzy inference model within the temperature accumulation effect correction submodule to obtain a fuzzy output result; and defuzzifying the fuzzy output result to obtain a temperature correction amount, and determining the corrected effective temperature as the sum of the original temperature and the temperature correction amount. The technical solution of this invention uses a continuous high-temperature intensity index to adaptively correct the temperature, accurately characterize the lag and cumulative effects, and significantly improve the accuracy and robustness of various load prediction models.

[0090] In one embodiment, the fuzzy output result includes a membership function, which comprises a function formed by each membership degree. If the temperature sequence includes discrete data, the defuzzification process of the fuzzy output result to obtain the temperature correction amount includes: S2701. The membership degrees included in the membership function are weighted and summed over the universe of discourse to obtain the weighted summation result.

[0091] In this embodiment, the weighted summation result can be the weighted summation result of discrete membership degrees. The weights can be candidate values ​​for the temperature correction amount.

[0092] Specifically, if the temperature sequence includes discrete data, the temperature correction amount can be determined based on the membership function composed of the discrete membership degrees. Specifically, the discrete membership degrees are first weighted and summed over the universe of discourse to obtain the weighted summation result.

[0093] S2702. Determine the sum of the membership degrees included in the membership function.

[0094] In this embodiment, the sum can be the sum of all membership degrees in the universe of discourse. The sum can reflect whether the fuzzy output results are concentrated.

[0095] Specifically, determine the sum of all membership degrees in the domain.

[0096] S2703. The ratio of the weighted summation result to the sum value is determined as the temperature correction amount.

[0097] Specifically, the ratio of the weighted sum to the sum value is determined as the temperature correction amount.

[0098] For example, the temperature correction amount can be expressed as follows: in, For the first The sampling point location can be a sampling point of the temperature correction amount on the universe of discourse, that is, the sampling point of the temperature correction amount. There are candidate values, and the universe contains a total of _____ candidate values. One sampling point. For the first The membership value corresponding to each sampling point.

[0099] In one example Figure 6This is a flowchart of a power load forecasting method provided in an embodiment of the present invention. First, in the scenario segmentation submodule, historical load data and meteorological data in the dataset are preprocessed. Corresponding features are extracted from the dataset through feature engineering, followed by clustering to determine the clustering results. Then, in the temperature cumulative effect correction submodule, the temperature sequence extracted from the dataset is input, the temperature sequence is fuzzified, and then fuzzy inference is performed according to preset rules. After defuzzification, a temperature correction amount is obtained. The temperature correction amount is summed with the original temperature to obtain the corrected effective temperature. Based on the clustering results, it is determined whether the set weather event to which the samples in the dataset belong includes a continuous high-temperature event. If so, the data is... The effective temperature after settling and correction is decomposed to obtain multiple intrinsic mode components and residuals (i.e., components). The sample entropy of each of the multiple intrinsic mode components and residuals is calculated. Based on the sample entropy, high-frequency components and low-frequency components are divided (i.e., high- and low-frequency component division). The high-frequency components are denoised and spliced ​​with meteorological features. The low-frequency components are spliced ​​with meteorological features to obtain spliced ​​high-frequency components and spliced ​​low-frequency components. The spliced ​​high-frequency components and spliced ​​low-frequency components are predicted separately to obtain the prediction results of high-frequency components and low-frequency components. The prediction results of high-frequency components and low-frequency components are fused to obtain the load prediction result.

[0100] The present invention will be described by way of example below: This invention aims to address key challenges in power load forecasting under frequent extreme weather conditions, including the difficulty in characterizing the temperature accumulation effect, sparse historical samples, and fluctuating load patterns. Traditional forecasting methods generally rely on a single model structure, lack fine-grained segmentation of extreme weather scenarios and dynamic temperature correction capabilities. Furthermore, under extreme weather conditions, load sequences exhibit strong non-stationarity and high-frequency abrupt changes, resulting in insufficient model capture of peak loads and significant prediction bias. Simultaneously, existing signal decomposition and deep learning fusion methods face challenges under extreme weather conditions, such as weak noise suppression, insufficient multi-scale feature fusion, and excessive model complexity, leading to poor robustness and unreliable real-time performance, failing to provide accurate and reliable technical support for power grid dispatching and risk control. Therefore, the core technical problem of this invention is to construct a short-term load forecasting method that integrates extreme weather scenario segmentation (implemented by a scenario segmentation submodule), dynamic correction of temperature accumulation effects (implemented by a temperature accumulation effect correction submodule), and adaptive signal decomposition and reconstruction (implemented by an adaptive decomposition prediction submodule). This method improves forecasting accuracy while enhancing the model's adaptability to extreme load fluctuations, providing effective decision-making basis for power grid safe operation and emergency dispatch.

[0101] With the intensification of global climate change and the advancement of new power system construction, the impact of extreme weather events on power load is becoming increasingly prominent. Load curves exhibit complex characteristics such as drastic fluctuations, strong non-stationarity, and peak value anomalies, posing severe challenges to real-time grid balancing and safe regulation. Against this backdrop, high-precision and robust short-term load forecasting methods have become key technologies supporting grid disaster prevention and dispatching and risk early warning. However, existing forecasting methods still face three major bottlenecks under extreme weather conditions: first, insufficient characterization of key physical mechanisms such as temperature accumulation effects and meteorological lag responses; second, the scarcity of historical extreme weather samples leads to insufficient model training and limited generalization performance; and third, the multi-scale and non-stationary characteristics of load sequences under extreme conditions make it difficult for traditional single models to simultaneously achieve trend fitting and peak value capture, severely impacting forecast accuracy and practical value.

[0102] In the identification of extreme weather scenarios, existing traditional clustering algorithms mostly rely on Euclidean distance metrics, which are insensitive to differences in load curve morphology and make it difficult to achieve fine-grained classification. In the modeling of temperature accumulation effects, although existing studies have considered the cumulative effect of temperature, they tightly couple the effect with the model structure and lack modular dynamic correction capabilities. In the signal decomposition and prediction fusion, although existing frameworks improve the modeling effect through decomposition and noise reduction, they still have shortcomings such as insufficient high-frequency noise suppression, rigid component reconstruction strategies, and excessive model complexity under extreme weather conditions, making it difficult to meet the real-time prediction requirements while ensuring accuracy.

[0103] Existing extreme weather load forecasting methods rely heavily on traditional clustering algorithms in the scene segmentation stage. While their Euclidean distance-based similarity metric is sensitive to changes in load curve amplitude, it struggles to capture the essential differences in temporal morphological structures. This results in coarse load pattern recognition and clustering results containing atypical curves under extreme weather conditions. Consequently, subsequent prediction models suffer from inaccurate scene segmentation and weak data-driven modeling foundations, significantly reducing their ability to characterize extreme load peaks and fluctuations. Therefore, this invention provides a hybrid clustering method for fine segmentation of extreme weather scenes. By introducing a shape distance metric (i.e., shape distance) and an adaptive cluster number determination mechanism (i.e., determining the optimal number of candidate clusters from the number of candidate clusters in the search interval), it achieves keen identification of load curve morphological features, effectively improving the interpretability and modeling quality of extreme scene classification.

[0104] Because existing methods for modeling the cumulative effect of temperature mostly employ static indicators or are tightly coupled with the prediction model, they lack flexible quantification of the dynamic correlation between the number of consecutive days of high temperatures, the cumulative intensity, and the predicted daily temperature. This leads to distortion of the load response pattern and decreased prediction accuracy when the temperature suddenly changes after a period of sustained high temperatures. Consequently, traditional correction strategies struggle to accurately reflect the lag relationship between perceived temperature and load growth, especially exhibiting insufficient model generalization ability under highly volatile extreme weather conditions. Therefore, this invention designs a fuzzy inference-based dynamic temperature correction method. By constructing a multi-input fuzzy rule base and performing defuzzification processing, it achieves modular and adaptive quantification of the cumulative effect of temperature, thereby enhancing the robustness and accuracy of various load prediction models under extreme weather conditions.

[0105] Traditional signal decomposition and prediction frameworks face challenges under extreme weather conditions, including weak high-frequency noise suppression, rigid component reconstruction strategies, and poor adaptability to multi-scale feature fusion. These issues result in insufficient handling of strong non-stationarity in load sequences, significant peak prediction lag, and substantial amplitude deviations. Consequently, a single model struggles to account for both long-term trends and short-term abrupt changes, leading to prediction intervals that deviate from actual fluctuations during extreme events such as cold waves and high temperatures, thus failing to provide reliable scheduling guidance. Therefore, this invention proposes a method based on adaptive signal decomposition and component reconstruction technology, coupled with a dual-branch prediction model. Through component partitioning and targeted noise suppression, it achieves a balanced optimization of load curve trend fitting and peak capture under extreme weather conditions, significantly improving prediction accuracy and robustness.

[0106] The short-term power load forecasting method under extreme weather conditions that considers the temperature accumulation effect includes three continuous processing modules (i.e., scenario partitioning submodule, temperature accumulation effect correction submodule, and adaptive decomposition forecasting submodule), which together constitute an integrated forecasting model (i.e., power load forecasting model).

[0107] First, a two-stage shape clustering method is used to perform on the load time series under extreme weather conditions by combining coarse clustering (i.e., determining the search interval first) and fine clustering (i.e., using the average silhouette coefficient of each sample in each candidate clustering result, and the number of candidate clusters corresponding to the largest average value among all the average values ​​as the final number of clusters). This method automatically determines the number of clusters while highlighting the similarity measurement of curve shapes, thus achieving a fine division of extreme weather load scenarios. The final output includes scenario category labels, corresponding clusters, and the centers of typical load curve clusters; these are used to characterize the mapping relationship between extreme weather characteristics and load change patterns, serving as the basis for scenario-specific modeling.

[0108] The devices corresponding to the scene segmentation sub-modules may include a data acquisition and storage module, a preprocessing and coarse clustering module, a fine clustering module, and a result output module. Each module is interconnected via a bus on a hardware circuit consisting of a processor and a memory to realize online identification and output of extreme weather load scenarios.

[0109] Secondly, the cumulative temperature effect of continuous high temperature on short-term power load is characterized by a fuzzy inference model. The number of days and intensity of high temperature are mapped into temperature correction values. The predicted daily temperature (i.e. the date corresponding to the original temperature in the temperature series) is dynamically corrected and decoupled from the load forecasting model.

[0110] The device corresponding to the temperature accumulation effect correction submodule may include a data acquisition module, an accumulated temperature calculation module, a fuzzy inference and defuzzification module, and a load prediction module. Each module is deployed on the processor, memory, and analog-to-digital conversion circuit. Data interaction is completed through the system bus, and the output is the temperature correction amount and the corrected effective temperature. The corrected effective temperature is used as one of the input features of the subsequent load prediction module (i.e., the adaptive decomposition prediction submodule).

[0111] Finally, through a two-stage structure of "adaptive signal decomposition (i.e., decomposing the input features into multiple intrinsic mode components and residuals) – component reconstruction (i.e., splicing the denoised high-frequency components and the low-frequency components with the corresponding meteorological features respectively) – self-attention – dual-branch prediction model (i.e., inputting the spliced ​​high-frequency components and the spliced ​​low-frequency components into different branches of the dual-branch prediction model respectively, with each branch outputting high-frequency component prediction results and low-frequency component prediction results respectively, and fusing the high-frequency component prediction results and low-frequency component prediction results through a self-attention method), multi-scale modeling and fine prediction of drastically fluctuating power load under extreme weather conditions are performed.

[0112] The adaptive decomposition prediction submodule can be comprised of a data acquisition and storage module, an adaptive signal decomposition module, a component reconstruction module, a self-attention-dual-branch prediction module, and a result output module, with these three parts sequentially connected. The first part divides historical load data under extreme weather conditions into scenarios, outputting scenario category labels, clusters, and typical load curve cluster centers for subsequent scenario identification or scenario-specific modeling. The second part corrects for the cumulative temperature effect on the predicted daily temperature, outputting the temperature correction amount and the corrected effective temperature, which constitute the input features for subsequent load prediction. The third part, based on historical load and meteorological data, combines the scenario division results from the first part with the corrected temperature features from the second part to decompose, reconstruct, and predict the load sequence, ultimately obtaining short-term power load prediction results and achieving high-precision prediction of short-term power load under extreme weather conditions.

[0113] This invention can automatically determine the number of clusters and identify differentiated patterns based on the shape characteristics of the load curve, effectively improving the identifiability and modeling foundation of extreme weather scenarios; it can adaptively correct the predicted daily temperature by integrating the number of consecutive days of high temperature and the cumulative intensity index, accurately characterizing the lag and cumulative effects, and significantly improving the accuracy and robustness of multiple load prediction models; it can divide the load sequence into high-frequency and low-frequency components according to complexity, and use a dual-branch self-attention temporal convolutional network for differentiated modeling, effectively balancing long-term trend tracking and peak fluctuation capture capabilities.

[0114] Example 3 Figure 7 This is a schematic diagram of the structure of a power load forecasting device provided in Embodiment 3 of the present invention. Figure 7 As shown, the device includes: The acquisition module 310 is used to acquire a dataset, which includes historical power load data and meteorological data under a set weather event, and the set weather event includes weather events that have an impact on the power load; The clustering result determination module 320 is used to determine the clustering result based on the dataset through the scenario division submodule in the power load prediction model. The clustering result indicates the mapping relationship between the characteristics of various set weather events and the corresponding load change patterns. Temperature correction module 330 is used to correct the temperature of the dataset through the temperature accumulation effect correction submodule in the power load prediction model, and determine the corrected effective temperature. Output module 340 is used to perform power load prediction based on the dataset, the effective temperature and the clustering results through the adaptive decomposition prediction submodule in the power load prediction model, and output the load prediction result, which indicates the prediction result of the power load under the set weather event.

[0115] The technical solution of this invention involves: an acquisition module acquiring a dataset, which includes historical power load data and meteorological data under a set weather event, wherein the set weather event includes weather events that affect power load; a clustering result determination module determining clustering results based on the dataset using a scenario segmentation submodule in a power load prediction model, wherein the clustering results indicate the mapping relationship between the characteristics of various set weather events and corresponding load change patterns; a temperature correction module correcting the dataset for temperature using a temperature accumulation effect correction submodule in a power load prediction model to determine the corrected effective temperature; and an output module performing power load prediction based on the dataset, the effective temperature, and the clustering results using an adaptive decomposition prediction submodule in a power load prediction model, outputting a load prediction result indicating the predicted power load under the set weather event. The technical solution of this invention determines the clustering results through a scenario segmentation submodule. The clustering results indicate the mapping relationship between the characteristics of a set weather event and the corresponding load change pattern, improving the scenario segmentation accuracy of the adaptive decomposition prediction submodule, laying the foundation for data-driven modeling, and improving the ability to characterize extreme load peaks and fluctuations under set weather events. A temperature accumulation effect correction submodule performs temperature correction to determine the corrected effective temperature, solving the problem of insufficient characterization of temperature accumulation effects in power load prediction models and improving the prediction accuracy of power load data. Finally, the adaptive decomposition prediction submodule outputs load prediction results, further improving the prediction accuracy of power load data under set weather events.

[0116] In one embodiment, the clustering result determination module 320 is specifically used for: The dataset is clustered by the scene segmentation submodule. The data in the dataset is grouped into multiple categories according to a preset distance threshold range, and the search range is determined according to the categories. The scenario segmentation submodule traverses the number of candidate clusters in the search interval. For each number of candidate clusters, it determines the candidate clustering result corresponding to each number of candidate clusters. It also determines the first average distance from each sample in each candidate cluster to samples in other clusters, and the second average distance from each sample in each cluster to other samples in the cluster. The scene segmentation submodule determines the contour coefficient of each sample in each of the candidate clustering results based on the first average distance and the second average distance. The average silhouette coefficient of each sample in each candidate clustering result is determined by the scene segmentation submodule. The number of candidate clusters corresponding to the largest average value among all the average values ​​is taken as the final cluster number, and the clustering result corresponding to the final cluster number is determined.

[0117] In one embodiment, the first average distance includes the average shape distance of each sample within a cluster to samples in other clusters in the clustering result; the second average distance includes the average shape distance of each sample within a cluster to other samples in the cluster group.

[0118] In one embodiment, the temperature correction module 330 further includes: An extraction unit is used to extract a temperature sequence from the dataset through a temperature accumulation effect correction submodule, the temperature sequence including multiple raw temperatures; The determining unit is used to determine the continuous high temperature intensity index of the temperature sequence based on a preset high temperature threshold and the temperature sequence through the temperature accumulation effect correction submodule. The fuzzification unit is used to perform fuzzification processing on the temperature sequence and the continuous high temperature intensity index respectively through the temperature accumulation effect correction submodule to obtain the corresponding membership degree; The fuzzy inference unit is used to input the membership degree into the fuzzy inference model in the temperature accumulation effect correction submodule through the temperature accumulation effect correction submodule to obtain the fuzzy output result; The defuzzification unit is used to defuzzify the fuzzy output result through the temperature accumulation effect correction submodule to obtain the temperature correction amount, and to determine the sum of the original temperature and the temperature correction amount as the corrected effective temperature.

[0119] In one embodiment, the defuzzification unit is specifically used for: The membership degrees included in the membership function are weighted and summed over the universe of discourse to obtain the weighted summation result; Determine the sum of the membership degrees included in the membership function; The ratio of the weighted sum to the sum value is determined as the temperature correction amount.

[0120] In one embodiment, the output module 340 further includes: The judgment unit is used to determine, based on the clustering results, whether the set weather event to which the samples in the dataset belong includes a continuous high temperature event, wherein the continuous high temperature event includes an event in which the temperature is higher than a set threshold for a continuous set duration; if so, the dataset and the effective temperature are used as input features, and the input features are decomposed into multiple intrinsic mode components and residuals. The partitioning unit is used to determine the sample entropy of the intrinsic mode components and the residuals through the adaptive decomposition prediction submodule, and to partition the intrinsic mode components and the residuals corresponding to each sample entropy into high-frequency components and low-frequency components according to the numerical relationship between each sample entropy and the sample entropy threshold. The fusion unit is used to determine the corresponding high-frequency component prediction result and low-frequency component prediction result respectively based on the high-frequency component and the low-frequency component through the adaptive decomposition prediction submodule, and fuse the high-frequency component prediction result and the low-frequency component prediction result to obtain the load prediction result.

[0121] In one embodiment, the fusion unit is specifically used for: The high-frequency components are denoised to obtain the denoised high-frequency components. The noise-reduced high-frequency component and the low-frequency component are respectively concatenated with the corresponding meteorological features to obtain the concatenated high-frequency component and the concatenated low-frequency component. The meteorological features include meteorological features extracted from the meteorological data. The spliced ​​high-frequency component and the spliced ​​low-frequency component are respectively input into different branches of the dual-branch prediction model, and each branch outputs the prediction results of the high-frequency component and the low-frequency component, respectively.

[0122] The power load forecasting device provided in this embodiment of the invention can execute the power load forecasting method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0123] Example 4 Figure 8 A schematic diagram of an electronic device that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0124] like Figure 8As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0125] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0126] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the methods proposed in this invention.

[0127] In some embodiments, the method proposed in this invention can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the method described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the method proposed in this invention by any other suitable means (e.g., by means of firmware).

[0128] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard parts (ASSPs), systems-on-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0129] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0130] In the context of this invention, a computer-readable storage medium stores computer instructions that are used to cause a processor to execute and implement the method provided by this invention.

[0131] The present invention also provides a computer program product comprising a computer program that, when executed by a processor, implements the method provided according to embodiments of the present invention.

[0132] Computer-readable storage media can be tangible media that may contain or store computer programs for use by or in conjunction with an instruction execution system, apparatus, or device. Computer-readable storage media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0133] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device for displaying information to the user, such as a cathode ray tube (CRT) or a liquid crystal display (LCD); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0134] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0135] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system. It addresses the shortcomings of traditional physical hosts and Virtual Private Server (VPS) services, such as high management difficulty and weak business scalability.

[0136] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0137] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A power load forecasting method characterized by, include: Obtain a dataset, which includes historical power load data and meteorological data under a set weather event, wherein the set weather event includes weather events that have an impact on power load; The scenario segmentation submodule in the power load forecasting model determines the clustering results based on the dataset. The clustering results indicate the mapping relationship between the characteristics of various set weather events and the corresponding load change patterns. The effective temperature after correction is determined by using the temperature accumulation effect correction submodule in the power load prediction model to correct the temperature of the dataset. The adaptive decomposition prediction submodule in the power load prediction model performs power load prediction based on the dataset, the effective temperature, and the clustering results, and outputs the load prediction results, which indicate the predicted power load under the set weather event.

2. The method of claim 1, wherein, The process of determining clustering results based on the dataset using the scenario segmentation submodule in the power load prediction model includes: The scene segmentation submodule categorizes the data in the dataset into multiple categories based on a preset distance threshold range, and determines the search range based on the categories. The scenario segmentation submodule traverses the number of candidate clusters in the search interval. For each number of candidate clusters, it determines the candidate clustering result corresponding to each number of candidate clusters. It also determines the first average distance from each sample in each candidate cluster to samples in other clusters, and the second average distance from each sample in each cluster to other samples in the cluster. The scene segmentation submodule determines the contour coefficient of each sample in each of the candidate clustering results based on the first average distance and the second average distance. The average silhouette coefficient of each sample in each candidate clustering result is determined by the scene segmentation submodule. The candidate cluster number corresponding to the largest average value among all the average values ​​is taken as the final cluster number, and the clustering result corresponding to the final cluster number is determined.

3. The method of claim 2, wherein, The first average distance includes the average shape distance of each sample within a cluster to samples in other clusters in the clustering result; the second average distance includes the average shape distance of each sample within a cluster to other samples in the cluster group.

4. The method of claim 1, wherein, The step of correcting the dataset for temperature using the temperature accumulation effect correction submodule in the power load forecasting module to determine the corrected effective temperature includes: Temperature sequences are extracted from the dataset by a temperature accumulation effect correction submodule, the temperature sequences including multiple raw temperatures; The temperature accumulation effect correction submodule determines the continuous high temperature intensity index of the temperature sequence based on a preset high temperature threshold and the temperature sequence. The temperature sequence and the continuous high temperature intensity index are fuzzified by the temperature accumulation effect correction submodule to obtain the corresponding membership degree. The membership degree is input into the fuzzy inference model in the temperature accumulation effect correction submodule to obtain the fuzzy output result. The fuzzy output result is defuzzified by the temperature accumulation effect correction submodule to obtain the temperature correction amount, and the sum of the original temperature and the temperature correction amount is determined as the corrected effective temperature.

5. The method of claim 4, wherein, The fuzzy output result includes a membership function, which comprises a function formed by each membership degree. If the temperature sequence includes discrete data, the defuzzification process of the fuzzy output result to obtain the temperature correction amount includes: The membership degrees included in the membership function are weighted and summed over the universe of discourse to obtain the weighted summation result; Determine the sum of the membership degrees included in the membership function; The ratio of the weighted sum to the sum value is determined as the temperature correction amount.

6. The method of claim 1, wherein, The adaptive decomposition prediction submodule in the power load prediction model performs power load prediction based on the dataset, the effective temperature, and the clustering results, and outputs load prediction results, including: The adaptive decomposition prediction submodule determines, based on the clustering results, whether the samples in the dataset belong to a set weather event that includes consecutive high-temperature events, where the consecutive high-temperature events include events where the temperature is higher than a set threshold for a continuous set duration; if so, the dataset and the effective temperature are used as input features, and the input features are decomposed into multiple intrinsic mode components and residuals. The intrinsic mode components and the residuals are determined by the adaptive decomposition prediction submodule. Based on the numerical relationship between each sample entropy and the sample entropy threshold, the intrinsic mode components and the residuals corresponding to each sample entropy are divided into high-frequency components and low-frequency components. The adaptive decomposition prediction submodule determines the corresponding high-frequency component prediction result and low-frequency component prediction result based on the high-frequency component and the low-frequency component, respectively, and then fuses the high-frequency component prediction result and the low-frequency component prediction result to obtain the load prediction result.

7. The method of claim 6, wherein, The step of determining the corresponding high-frequency component prediction result and low-frequency component prediction result based on the high-frequency component and low-frequency component respectively includes: The high-frequency components are denoised to obtain the denoised high-frequency components. The noise-reduced high-frequency component and the low-frequency component are respectively concatenated with the corresponding meteorological features to obtain the concatenated high-frequency component and the concatenated low-frequency component. The meteorological features include meteorological features extracted from the meteorological data. The spliced ​​high-frequency component and the spliced ​​low-frequency component are respectively input into different branches of the dual-branch prediction model, and each branch outputs the prediction results of the high-frequency component and the low-frequency component, respectively.

8. An electric power load forecasting device characterized by comprising: include: The acquisition module is used to acquire a dataset, which includes historical power load data and meteorological data under a set weather event, and the set weather event includes weather events that have an impact on the power load; The clustering result determination module is used to determine the clustering result based on the dataset through the scenario segmentation submodule in the power load prediction model. The clustering result indicates the mapping relationship between the characteristics of various set weather events and the corresponding load change patterns. The temperature correction module is used to correct the temperature of the dataset through the temperature accumulation effect correction submodule in the power load prediction model, and to determine the corrected effective temperature. The output module is used to perform power load prediction based on the dataset, the effective temperature and the clustering results through the adaptive decomposition prediction submodule in the power load prediction model, and output the load prediction result, which indicates the prediction result of the power load under the set weather event.

9. An electronic device, comprising: The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A computer program product, characterised in that, The computer program product includes a computer program that, when executed by a processor, implements the method according to any one of claims 1-7.

Images