Power load prediction method and related device based on multi-representation adaptive alignment

Abstract

This invention discloses a power load forecasting method and related apparatus based on multi-representation adaptive alignment, belonging to the field of power load forecasting technology. The method first acquires load observation data from multiple sub-regions within a target area over a continuous time interval and performs preprocessing. Time-series representations are constructed in parallel from the preprocessed data, and then fused features are constructed. A multi-channel adaptive alignment mechanism is used to align and model the time-series representations and fused features to obtain aligned features. The fused features and aligned features are then residually fused to obtain modeled features. The modeled features are input into a prediction mapping function to obtain the power load forecasting results for each sub-region within the target area. This invention improves the accuracy, stability, and engineering deployability of load forecasting by jointly characterizing load evolution patterns in different time feature spaces and introducing structural difference constraints to coordinate the common and individual characteristics of loads between regions.

Description

Technical Field

[0001] This invention belongs to the field of power load forecasting technology, and relates to a power load forecasting method and related apparatus based on multi-representation adaptive alignment. Background Technology

[0002] With the large-scale grid integration of new energy sources such as wind and solar power, the deepening of power market reforms, and the increasing abundance of demand-side adjustable resources, the composition and dynamic behavior of regional power loads are becoming increasingly complex. Their time series exhibit significant characteristics such as non-stationarity, nonlinearity, and multi-timescale coupling. Specifically, load sequences simultaneously contain intraday periodicity, weekly periodicity, long-term trends, and random disturbances caused by weather, holidays, and unforeseen events. Different forecasting steps (such as ultra-short-term, short-term, and medium-term) emphasize different aspects of load evolution patterns; short-term forecasts focus more on local fluctuation patterns, while medium- and long-term forecasts rely more on overall structural stability and periodic patterns.

[0003] To address these challenges, load forecasting technology has evolved from traditional statistical methods (such as time series analysis and regression models) to artificial intelligence methods based on machine learning and deep learning, significantly improving forecast accuracy. However, when facing complex forecasting scenarios involving multiple regions and time scales, existing technologies still suffer from several common shortcomings that urgently need to be addressed, mainly in the following aspects: Most methods only model the spatial representation in a single time period, making it difficult to simultaneously characterize trend structures and local disturbances; the structural differences between multivariate loads lack explicit constraints, making the models susceptible to scale and phase shifts; the forecasting models lack adaptability to forecast step sizes, resulting in unstable performance at different forecast scales; and the mathematical structure lacks interpretability, hindering engineering deployment and parameter review.

[0004] In summary, there is an urgent need in this field for a new power load forecasting method to jointly characterize the complex evolution of regional loads from multiple perspectives and scales, effectively coordinate the common and individual characteristics among loads in multiple regions, and adaptively match the forecasting needs of different time scales. Ultimately, this method can improve forecasting accuracy while enhancing the stability, interpretability, and engineering practical value of the model. Summary of the Invention

[0005] The purpose of this invention is to provide a power load forecasting method and related apparatus based on multi-representation adaptive alignment, so as to solve the technical problem of low accuracy of load forecasting results under different forecasting step sizes and complex operating scenarios in the prior art.

[0006] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, the present invention provides a power load forecasting method based on multi-representation adaptive alignment, comprising the following steps: Acquire load observation data of multiple sub-regions within the target area over a continuous time interval, and perform preprocessing; Temporal representations are constructed in parallel from the preprocessed data, and then fused features are constructed. By utilizing a multi-channel adaptive alignment mechanism, alignment modeling is performed on temporal representations and fused features to obtain aligned features; The fused features and aligned features are then residually fused to obtain the modeling features; The modeling features are input into the prediction mapping function to obtain the power load prediction results for each sub-region within the target area.

[0007] Furthermore, the step of acquiring load observation data of multiple sub-regions within the target area over a continuous time interval and performing preprocessing specifically includes: Obtain load observation data from multiple sub-regions within the target area over a continuous time interval to construct a load data tensor. ,in, This indicates the number of load sub-regions under the power grid. T Indicates the number of historical observation points; right Perform outlier correction, missing value imputation, and sliding standardization. A time-location-related standardized mapping is introduced to normalize each load observation data point. The specific formula is as follows:

[0008] in, Indicates the first The region in time location Local mean in the vicinity, This represents the corresponding local standard deviation. This is the normalized load value; Indicates the first The region in the first The actual load value at each time point.

[0009] Furthermore, the temporal representation includes original temporal evolution representation, periodic structure representation, and statistical semantic representation; The expression for the original time evolution representation is:

[0010] in, This represents the characteristic representation of the load in the time evolution space. For evolutionary mapping functions; The expression representing the periodic structure is:

[0011] in, This represents the characteristic representation of the load in the structural association space. It is a periodic mapping function; The expression for the statistical semantic representation is:

[0012] in, Used to provide supplementary characterization of the overall operating status and behavior patterns of the load; This is a statistical semantic mapping function.

[0013] Furthermore, the step of constructing the fusion feature specifically includes: Introducing prediction step size And construct the step size adjustment coefficient The importance of different representations is adaptively adjusted based on the prediction step size to construct fused features:

[0014] In the formula, This indicates the fusion feature.

[0015] Furthermore, the step of using a multi-channel adaptive alignment mechanism to align and model temporal representations and fused features to obtain aligned features specifically includes: A multi-channel adaptive alignment mechanism is introduced to model the alignment of the fused features:

[0016] in, Indicates the first One of the corresponding channels, Indicates the first Alignment mappings, For the first Alignment results for each channel; The output of each aligned channel is weighted Perform adaptive combination:

[0017] in, Indicates the first The importance weight of each alignment channel in the final alignment result; This indicates alignment features.

[0018] Furthermore, the step of residual fusion of the fused features and the aligned features to obtain the modeling features specifically includes: By introducing structure-preserving residual coefficients, the fused features and aligned features are residually fused to obtain the modeling features:

[0019] in, To finally model the feature matrix, ∈[0,1] represents the structurally preserved residual coefficients.

[0020] Furthermore, the step of inputting the modeling features into the prediction mapping function to obtain the power load prediction results for each sub-region within the target area specifically includes: Modeling features Input prediction mapping function ( Generate future load forecast results:

[0021] in, This indicates that each region will be in the future Load forecast values ​​within each time step.

[0022] Secondly, the present invention provides a power load forecasting system based on multi-representation adaptive alignment, comprising: The preprocessing module is used to acquire load observation data of multiple sub-regions within the target area over a continuous time interval and to perform preprocessing. The feature fusion module is used to construct temporal representations of the preprocessed data in parallel, and then construct fused features. The adaptive alignment module is used to perform alignment modeling on temporal representations and fused features using a multi-channel adaptive alignment mechanism to obtain aligned features. The residual fusion module is used to perform residual fusion of the fused features and the aligned features to obtain the modeling features; The prediction output module is used to input the modeling features into the prediction mapping function to obtain the power load prediction results of each sub-region within the target area.

[0023] Thirdly, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the power load forecasting method based on multi-representation adaptive alignment as described above.

[0024] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the power load forecasting method based on multi-representation adaptive alignment.

[0025] Compared with the prior art, the present invention has the following beneficial effects: This invention discloses a power load forecasting method and related apparatus based on multi-representation adaptive alignment. First, by jointly modeling the load sequence using three complementary representations—time evolution, periodic structure, and statistical semantics—the system can more comprehensively depict the complex internal structure and change patterns of the load from multiple dimensions (trend, period, macroscopic behavior), solving the fundamental problem of insufficient modeling in a single representation space, thus significantly improving the model's representational ability and prediction accuracy. Second, the proposed prediction step-size-aware fusion mechanism can dynamically adjust the dependence weights on trend and periodic features according to different prediction tasks such as short-term and medium-term, enabling the model to adapt to prediction needs at different time scales and effectively improving the model's stability and generalization performance under different prediction step sizes. Finally, the core multi-channel adaptive alignment and structural difference constraint method not only effectively coordinates the commonalities and differences between loads in different regions at the feature level, weakening modeling biases caused by amplitude and phase inconsistencies, but also retains residual coefficients through structure preservation, promoting inter-regional information complementarity while avoiding excessive smoothing and loss of individual features. Overall, it enhances the robustness and interpretability of multi-regional collaborative prediction, providing a clear, stable, and adjustable technical path for the engineering deployment and integrated application of the model in real-world complex scenarios such as provincial power grids. Attached Figure Description

[0026] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the system of the present invention; Figure 3 This is a schematic diagram of the computer device structure of the present invention. Detailed Implementation

[0028] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present application.

[0029] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that the embodiments of this application 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 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.

[0030] It should be understood that in the embodiments of this application, "at least one" means one or more, and "more than one" means two or more. "And / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. The character " / " generally indicates that the related objects before and after it are in an "or" relationship. "Contains A, B and / or C" means containing any one, two, or three of A, B, and C.

[0031] It should be understood that in the embodiments of this application, "B corresponding to A", "B corresponding to A", "A corresponds to B" or "B corresponds to A" means that B is associated with A, and B can be determined based on A. Determining B based on A does not mean that B is determined solely based on A; B can also be determined based on A and / or other information.

[0032] This invention provides a power load forecasting method based on multi-representation adaptive alignment, which can be executed by an electronic device, such as a terminal or server. The terminal can be a smartphone, tablet, laptop, or other similar device. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. It is understood that this invention does not limit the specific entity executing this power load forecasting method based on multi-representation adaptive alignment.

[0033] The technical solution of this application will be described in detail below through specific embodiments. It should be noted that the following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments described below are used to explain the technical solution of this application and are not intended to limit actual use.

[0034] See Figure 1 This invention discloses a power load forecasting method based on multi-representation adaptive alignment, comprising the following steps: S1: Obtain load observation data of multiple sub-regions within the target area over a continuous time interval, and perform preprocessing; Obtain load observation data of the provincial power grid within a continuous time interval to construct a load data tensor: ,in, This indicates the number of load sub-regions under the power grid. Indicates the number of historical observation points. Indicates the first The region in the first The actual load value at each time point.

[0035] right Outlier correction, missing value imputation, and sliding standardization are performed.

[0036] To mitigate the non-stationarity of the load series, a time-location-dependent normalization mapping is introduced, and the following is performed on each load observation data:

[0037] in, Indicates the first The region in time location Local mean in the vicinity, This represents the corresponding local standard deviation. This is the normalized load value. This processing is used to mitigate the impact of long-term drift and scale changes on subsequent modeling.

[0038] S2, constructs temporal representations of the preprocessed data in parallel, and then constructs fused features; S201, Construction of Multi-Representation Temporal Mapping For the same normalized load sequence Three complementary temporal representations are constructed in parallel: (1) Primitive temporal evolution representation Define the time evolution mapping function Projecting continuous-time loads onto a low-dimensional time feature space:

[0039] in, This represents the characteristics of load in the time evolution space, used to reflect the trend and stage-based changes in load over time. As a time feature dimension, This represents the overall temporal evolution characteristics of region r.

[0040] It is a parameterized mapping module whose internal parameters are automatically determined during the system training phase by learning from historical load data, rather than being preset by humans to a fixed function form.

[0041] (2) Characterization of periodic structure Considering that electricity loads generally exhibit repetitive variations such as daily and weekly cycles, a periodic mapping function is introduced to highlight the periodicity and oscillation structure in the load sequence. Mapping the normalized load sequence to the periodic feature space yields:

[0042] in, This characterization represents the load characteristics in the structural correlation space, reflecting the differential change patterns under different regions, time periods, or load composition conditions. This characterization focuses on depicting the repetitive patterns and structural fluctuations of the load at different time scales, and is used to compensate for the shortcomings of simple time evolution characterization in depicting periodic information.

[0043] It is a parameterized mapping module whose internal parameters are automatically determined during the system training phase by learning from historical load data, rather than being preset by humans to a fixed function form.

[0044] (3) Statistical semantic representation To describe load characteristics from an overall behavioral perspective, a statistical descriptive vector is constructed based on a normalized load sequence. This descriptive vector includes statistical information such as load mean, variation amplitude, peak-valley characteristics, and time labels, and is mapped using a statistical semantic mapping function. get:

[0045] in, It is used to provide supplementary characterization of the overall operating status and behavior patterns of the load, making the model more robust when facing abnormal fluctuations or structural changes.

[0046] It is a parameterized mapping module whose internal parameters are automatically determined during the system training phase by learning from historical load data, rather than being preset by humans to a fixed function form.

[0047] It should be noted that the specific implementation forms of the time evolution mapping function, periodic mapping function, and statistical semantic mapping function are not uniquely limited. Their internal mapping structures can be replaced with parameterized mapping structures based on linear mapping, nonlinear mapping, or a combination of both, depending on the characteristics of the load data, to adapt to the differences in load variation trends, periodic characteristics, or statistical distributions in different regions. The periodic structure representation can be replaced with feature representations constructed based on different time scale decomposition methods, such as using multi-scale time windows, hierarchical time period aggregation, or other time structure decomposition methods, to enhance the ability to characterize the load variation patterns of different period lengths.

[0048] S202, Predictive Step Size Aware Adaptive Representation Fusion To address the issue of varying time scale sensitivities across different prediction tasks, a prediction step size is introduced. And construct the step size adjustment coefficient This is used to adaptively adjust the importance of different representations based on the prediction step size, and construct fused features:

[0049] When the prediction step size is short Larger values ​​are chosen to enhance the focus on short-term evolutionary features; when the prediction step size is longer, the weight of periodic structure representation in the fusion result is correspondingly increased to enhance the ability to characterize long-term patterns. In this way, the fusion features can adaptively match the prediction needs of different time scales.

[0050] Preferably, the step size adjustment coefficient can be replaced by an adaptive parameter that is dynamically adjusted according to the prediction task, or updated based on historical prediction error statistics, to further improve the adaptability of the fused features under different prediction step size conditions. The number of alignment channels and the mapping form of each alignment channel can be adjusted according to the application scenario. The multi-channel alignment mapping can be replaced by single-channel alignment, group alignment, or hierarchical alignment to adapt to application scenarios with different regional scales or large differences in regional load correlation.

[0051] S3, Multi-channel adaptive alignment modeling: Using a multi-channel adaptive alignment mechanism, alignment modeling is performed on temporal representations and fused features to obtain aligned features; Because loads in different regions vary in amplitude, rate of change, and phase position, direct fusion may lead to feature mismatch. Therefore, a multi-channel adaptive alignment mechanism is introduced to model the alignment of the fused features:

[0052] in, Indicates the first One of the corresponding channels, Indicates the first An alignment map is used to structurally align load features from different angles. For the first Alignment results for each channel.

[0053] The output of each aligned channel is weighted Perform adaptive combination:

[0054] in Indicates the first The importance weight of each alignment channel in the final alignment result. This indicates alignment features.

[0055] S4, Structural Difference Constraints and Residual Fusion: The fused features and aligned features are residually fused to obtain the modeling features; After completing the alignment modeling, to avoid the loss of individual features in the region due to over-alignment, a structure-preserving residual coefficient is introduced. ∈[0,1], construct the final modeling features used for prediction:

[0056] in, To finally model the feature matrix, Used to control the The degree to which each region retains its original structural features during the fusion process is considered, thus achieving a balance between modeling regional commonalities and preserving individual characteristics.

[0057] It should be noted that the setting method for the structural retention residual coefficient can be replaced by a method that dynamically adjusts it based on the region's historical stability, load fluctuation amplitude, or prediction error feedback, in order to achieve a better balance between maintaining the individual characteristics of the region and enhancing the modeling of the region's common features.

[0058] S5. Input the modeling features into the prediction mapping function to obtain the power load prediction results for each sub-region within the target area.

[0059] The final features obtained in step S4 Input prediction mapping function ( This generates the future load forecast results for the provincial power grid.

[0060] in, This indicates that each region will be in the future Load forecast values ​​within each time step.

[0061] It should be noted that, in the load forecasting output stage, the specific implementation of the forecasting mapping function can be replaced according to the forecasting accuracy and computational resource requirements. For example, parameterized mapping structures of different complexities can be used to meet different application needs such as real-time forecasting or offline analysis. At the system application level, the method and system described in this invention are not only applicable to provincial power grid load forecasting, but can also be extended to municipal load forecasting, regional integrated energy load forecasting, power demand-side analysis, or other energy load forecasting scenarios with time series characteristics.

[0062] See Figure 2 This invention discloses a power load forecasting system based on multi-representation adaptive alignment, comprising a preprocessing module, a feature fusion module, an adaptive alignment module, a residual fusion module, and a forecast output module. The preprocessing module acquires and preprocesses load observation data from multiple sub-regions within a continuous time interval within a target area. The feature fusion module constructs time-series representations in parallel from the preprocessed data, thereby constructing fused features. The adaptive alignment module uses a multi-channel adaptive alignment mechanism to align and model the time-series representations and fused features to obtain aligned features. The residual fusion module performs residual fusion of the fused features and aligned features to obtain modeled features. The forecast output module inputs the modeled features into a forecast mapping function to obtain the power load forecasting results for each sub-region within the target area. This invention improves the accuracy, stability, and engineering deployability of load forecasting under different forecast step sizes and complex operating scenarios by jointly characterizing load evolution patterns in different time feature spaces and introducing structural difference constraints to coordinate the common and individual characteristics of loads between regions.

[0063] Example: This embodiment uses a province as an example to illustrate a power load forecasting method based on multi-representation adaptive alignment, which includes the following implementation steps: (1) Data sources and preprocessing: Historical load data from a provincial power grid over three consecutive years was selected as the implementation data source. The load data sampling interval was 15 minutes, and time stamp information related to load changes was also acquired, including date type and time period identifier. According to the provincial power grid dispatching and statistical standards, the province's load was divided into 10 regional nodes, with each regional node corresponding to a regional load time series.

[0064] In the data preprocessing stage, outliers in the original load data are first corrected, and a small number of missing data are imputed. Then, following the time-dependent normalization method in this invention, time-location-based sliding normalization is performed on the load sequences of each region to eliminate the influence of differences in load scale across different regions and long-term trend drift. After processing, a 10-dimensional multivariate load time-series matrix is ​​constructed, which serves as input data for subsequent multi-representation modeling.

[0065] (2) Model training: Historical load data from the previous two years were selected as training samples. Following the technical solution described in this invention, multi-representation construction, adaptive fusion, and alignment modeling were sequentially completed. The specific implementation process is as follows: 1) Multi-representation construction: Based on the normalized load time series data, time evolution representation, periodic structure representation and statistical semantic representation are constructed respectively to characterize the load characteristics from different perspectives such as the trend characteristics, periodicity and overall behavior patterns of load changes over time.

[0066] 2) Time window and feature dimension settings: The historical input time window length is set to 96 time steps, corresponding to 24 hours of load data, and the unified feature representation dimension is set to 64 to balance the model's expressive power and computational efficiency.

[0067] 3) Prediction step size perception fusion and alignment: For the 24-hour load forecasting task, a prediction step size perception mechanism is introduced to perform weighted fusion of time evolution representation and periodic structure representation. Combined with statistical semantic representation, the differences in amplitude level and change rhythm of load in different regions are reduced through multi-channel adaptive alignment. At the same time, a structure preservation mechanism is introduced during the fusion process to maintain the individual characteristics of load in each region.

[0068] 4) Model parameter learning: During the training process, the mapping parameters are learned through iterative optimization, so that the model gradually converges on the training data until the prediction error stabilizes.

[0069] (3) Prediction results: Historical load data from the third year were selected as test samples to perform rolling forecasts of the provincial power grid's load for the next 24 hours. The prediction results show that, under the same data conditions, the method described in this invention has a significantly lower average prediction error in the 24-hour forecasting task than the traditional linear forecasting method, and exhibits good stability under different regions and load levels.

[0070] In one embodiment of the invention, see [link to embodiment]. Figure 3A computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions from the computer storage medium to achieve a corresponding method flow or function. The processor described in this embodiment can be used in the operation of a power load forecasting method based on multi-representation adaptive alignment.

[0071] This invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the power load forecasting method based on multi-representation adaptive alignment in the above embodiments.

[0072] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0073] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0074] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0075] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A power load forecasting method based on multi-representation adaptive alignment, characterized in that, Includes the following steps: Acquire load observation data of multiple sub-regions within the target area over a continuous time interval, and perform preprocessing; Temporal representations are constructed in parallel from the preprocessed data, and then fused features are constructed. By utilizing a multi-channel adaptive alignment mechanism, alignment modeling is performed on temporal representations and fused features to obtain aligned features; The fused features and aligned features are then residually fused to obtain the modeling features; The modeling features are input into the prediction mapping function to obtain the power load prediction results for each sub-region within the target area. 2.The power load forecasting method based on multi-representation adaptive alignment according to claim 1, wherein, The step of acquiring load observation data of multiple sub-regions within the target area over a continuous time interval and performing preprocessing specifically includes: Obtaining load observation data of a plurality of sub-regions in the target region in a continuous time interval to form a load data tensor wherein, represents the number of load sub-regions under the power grid, T represents the number of historical observation time points; For anomaly value correction, missing value interpolation and sliding standardization processing; A time-location-related standardized mapping is introduced to normalize each load observation data point. The specific formula is as follows: in, Indicates the first The region in time location Local mean in the vicinity, This represents the corresponding local standard deviation. This is the normalized load value; Indicates the first The region in the first The actual load value at each time point.

3. The power load forecasting method based on multi-representation adaptive alignment according to claim 1, characterized in that, The temporal representation includes original temporal evolution representation, periodic structure representation, and statistical semantic representation; The expression for the original time evolution representation is: in, This represents the characteristic representation of the load in the time evolution space. For evolutionary mapping functions; The expression representing the periodic structure is: in, This represents the characteristic representation of the load in the structural association space. It is a periodic mapping function; The expression for the statistical semantic representation is: in, Used to provide supplementary characterization of the overall operating status and behavior patterns of the load; This is a statistical semantic mapping function.

4. The power load forecasting method based on multi-representation adaptive alignment according to claim 1, characterized in that, The steps for constructing the fusion features specifically include: Introducing prediction step size And construct the step size adjustment coefficient The importance of different representations is adaptively adjusted based on the prediction step size to construct fused features: In the formula, This indicates the fusion feature.

5. The power load forecasting method based on multi-representation adaptive alignment according to claim 1, characterized in that, The step of using a multi-channel adaptive alignment mechanism to align and model temporal representations and fused features to obtain aligned features specifically includes: A multi-channel adaptive alignment mechanism is introduced to model the alignment of the fused features: in, Indicates the first One of the corresponding channels, Indicates the first Alignment mappings, For the first Alignment results for each channel; The output of each aligned channel is weighted Perform adaptive combination: in, Indicates the first The importance weight of each alignment channel in the final alignment result; This indicates alignment features.

6. The power load forecasting method based on multi-representation adaptive alignment according to claim 1, characterized in that, The step of residual fusion of the fused features and the aligned features to obtain the modeling features specifically includes: By introducing structure-preserving residual coefficients, the fused features and aligned features are residually fused to obtain the modeling features: in, To finally model the feature matrix, ∈[0,1] represents the structurally preserved residual coefficients.

7. The power load forecasting method based on multi-representation adaptive alignment according to claim 1, characterized in that, The step of inputting the modeling features into the prediction mapping function to obtain the power load prediction results for each sub-region within the target area specifically includes: Modeling features Input prediction mapping function ( Generate future load forecast results: in, This indicates that each region will be in the future Load forecast values ​​within each time step.

8. A power load forecasting system based on multi-representation adaptive alignment, characterized in that, include: The preprocessing module is used to acquire load observation data of multiple sub-regions within the target area over a continuous time interval and to perform preprocessing. The feature fusion module is used to construct temporal representations of the preprocessed data in parallel, and then construct fused features. The adaptive alignment module is used to perform alignment modeling on temporal representations and fused features using a multi-channel adaptive alignment mechanism to obtain aligned features. The residual fusion module is used to perform residual fusion of the fused features and the aligned features to obtain the modeling features; The prediction output module is used to input the modeling features into the prediction mapping function to obtain the power load prediction results of each sub-region within the target area.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the power load forecasting method based on multi-representation adaptive alignment as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the power load forecasting method based on multi-representation adaptive alignment as described in any one of claims 1-7.

Images