Electricity Demand Prediction Using Area Feature Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing electricity-demand prediction methods face challenges in accurately predicting electricity demand at charging facilities due to limited record data, particularly when analyzing group behavior trends in specific areas, as the number of probe cars may be insufficient to gather considerable data.
Innovation Solution
An electricity-demand prediction device that includes an area feature value prediction unit, an individual model acquisition unit, and a demand prediction computation unit, which uses predicted area feature values with high periodic regularity to generate individual models for accurate demand forecasting, incorporating factors like vehicle density, average speed, and average charging rates, and reflects user decision-making features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If group behavior trends in a specific area are statistically analyzed to construct models for reproducing features of all group behavior trends, then prediction accuracy is improved, but a large amount of record data is required which cannot be acquired when the number of probe cars is limited
Solution Approach 1:
The patent segments the prediction model into two distinct components: (1) an area feature value prediction model that captures group behavior trends at the area level, and (2) individual model parameters that capture user-specific characteristics. This segmentation allows the system to utilize limited probe car data for individual parameters while using area-level statistical data for group behavior patterns, thereby achieving high prediction accuracy without requiring extensive individual record data.
Solution Approach 2:
The patent introduces area feature values (such as area vehicle density, area average speed, and area average charging rate) as intermediary variables that bridge individual user behavior and group behavior trends. These area-level features serve as mediators that can be predicted with high accuracy from limited probe car data and then used to inform individual prediction models, effectively transferring information from the group level to the individual level without requiring direct observation of all individual behaviors.
2Measurement precision
If a large amount of record data is provided to construct models for reproducing group behavior trends, then prediction accuracy is improved, but the complexity of data collection and processing increases
Solution Approach 1:
The patent segments the data requirements into area-level aggregated data (which can be obtained from existing traffic and charging infrastructure data sources) and individual-level parameters (which are obtained from limited probe car data). This segmentation eliminates the need for collecting and processing large amounts of individual record data, thereby reducing data collection and processing complexity while maintaining prediction accuracy.
Solution Approach 2:
The area feature value prediction model is trained to automatically generate area-level feature values that capture group behavior trends. Once trained, this model self-generates the necessary area-level inputs for individual prediction models without requiring manual data collection or complex processing of raw probe car data, thereby reducing operational complexity.
3Measurement precision
If individual models are constructed for each specific vehicle using area feature values as input, then prediction accuracy for individual vehicles is improved, but the computational complexity increases
Solution Approach 1:
The patent performs preliminary action by pre-training the area feature value prediction model and pre-determining individual model parameters using available probe car data. This preliminary processing creates a ready-to-use framework where individual vehicle predictions can be made by simply inputting current area feature values into the pre-established models, significantly reducing the computational complexity of real-time predictions while maintaining high accuracy.
Solution Approach 2:
The patent changes the parameters of individual models based on area feature values rather than retraining entire models for each vehicle. By adjusting model parameters (such as scaling factors or offset values) based on area-level conditions, the system achieves adaptive individual predictions with minimal computational overhead compared to full model retraining.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
This electricity-demand prediction device (100) is provided with an area-feature-value prediction unit (114), an individual-model acquisition unit (112), and a demand-prediction computation unit (113). Measured data relating to vehicles is input to the area-feature-value prediction unit (114), and for each of a plurality of areas, the area-feature-value prediction unit (114) predicts an area feature value for a feature relating to vehicles associated with the area in question. The individual-model acquisition unit (112) acquires individual models for specific vehicles. Said individual models take area feature values as input and output electricity-demand values for the specific vehicles at a specific charging facility. The demand-prediction computation unit (113) inputs the predicted area values to the individual models to compute a predicted electricity demand for the specific vehicles corresponding to said individual models at the aforementioned specific charging facility.