Electric Load Forecasting Using Variational Mode Decomposition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current electric load forecasting methods, such as those based on data periodicity and neural networks, struggle to accurately predict load trends due to limited feature extraction from single-feature load data, leading to unsatisfactory results, especially in complex scenarios.
Innovation Solution
The method involves acquiring historical load data, replacing outliers, generating a load sequence through variational mode decomposition, and inputting intrinsic mode components and residuals into forecasting models to determine load values, allowing for more dimensional feature extraction and improved forecasting accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If simple forecasting methods based on data periodicity are used, then the method is easy to implement, but the forecasting accuracy deteriorates in complex scenarios
Solution Approach 1:
The patent applies segmentation by decomposing the original load sequence into multiple intrinsic mode functions (IMFs) and a residual component through variational mode decomposition. Each IMF captures different oscillatory modes at different frequency bands, allowing the forecasting model to process multiple segmented features simultaneously, thereby improving forecasting accuracy while maintaining implementation feasibility
Solution Approach 2:
The patent transforms the single-dimensional time series data into multi-dimensional feature space by extracting multiple IMFs with different frequency characteristics. This dimensional transformation enables the model to capture complex patterns from multiple perspectives, resolving the contradiction between simplicity and accuracy
2Adaptability or versatility
If neural network methods are used to solve nonlinear problems, then the capability to handle nonlinear relationships is improved, but the feature extraction capability deteriorates due to single-feature input
Solution Approach 1:
The patent segments the single load sequence into multiple IMFs, each representing different frequency components and oscillatory patterns. This segmentation preserves rich feature information that would be lost in single-feature input, while the neural network maintains its nonlinear processing capability by taking these multiple segmented features as inputs
Solution Approach 2:
The patent changes the parameter dimensionality by transforming single load values into multiple IMF parameters with different frequency and amplitude characteristics. This parameter transformation enriches the input features without losing the neural network's ability to handle nonlinear relationships
Data Source
AI summary
A method and a device for forecasting an electric load, an electronic device, and a computer-readable storage medium are provided. The method includes: acquiring historical load data prior to a forecast date; generating a load sequence based on the historical load data; performing variational mode decomposition on the load sequence, to obtain multiple intrinsic mode components and a residual that are corresponding to the load sequence; and inputting the multiple intrinsic mode components and the residual into respective forecasting models, and determining a load value on the forecast date based on forecasting results of all the forecasting models.