Ensemble Prediction Model for Electricity Demand Adaptation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current electricity demand prediction models are difficult to apply universally due to varying demander types and frequent changes in demander groups, requiring extensive manual tuning and relearning, which is labor-intensive and inefficient.

Innovation Solution

An information processing apparatus that generates multiple prediction models based on various conditions and weights them to create an ensemble prediction model, allowing for accurate predictions with minimal tuning and adaptability to changes in demander groups.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If a single prediction model is used for all demanders, then device complexity is reduced, but prediction accuracy deteriorates due to varying demander types

Engineering Contradiction:
Improveprediction model structureVSAvoidprediction accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent segments the demander population into multiple clusters based on similar consumption patterns. Instead of using a single prediction model for all demanders, the system creates separate prediction models for each cluster, allowing each model to be optimized for specific demander types while maintaining overall system manageability.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent dynamically adjusts model parameters based on the identified cluster characteristics. By changing parameters such as model complexity, feature selection, and weighting schemes according to the specific cluster being predicted, the system achieves high accuracy for diverse demander types without requiring completely different model structures.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If manual tuning is performed for each demander group, then prediction accuracy is improved, but productivity deteriorates due to enormous daily workload

Engineering Contradiction:
Improveprediction accuracyVSAvoidwork efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent implements automated cluster identification and model selection algorithms that operate without manual intervention. The system automatically analyzes consumption patterns, identifies appropriate clusters, selects suitable prediction models for each cluster, and adjusts parameters dynamically, eliminating the need for manual tuning while maintaining high prediction accuracy.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system automatically adjusts model parameters based on real-time data and cluster characteristics, replacing manual parameter tuning with automated parameter optimization algorithms that adapt to changing demander group compositions without requiring human expertise or time investment.

Inventive Principle:
Principle #35Parameter changes

3Adaptability or versatility

If prediction models are frequently updated to adapt to changing demander groups, then adaptability is improved, but loss of time increases due to relearning requirements

Engineering Contradiction:
Improveadaptability to demander group changesVSAvoidrelearning time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent implements dynamic clustering that automatically adapts to changing demander group compositions. As new demanders join or leave the system, the clustering algorithm dynamically reorganizes clusters and adjusts model assignments without requiring complete relearning, allowing the system to maintain adaptability while minimizing relearning time through incremental updates.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3965049A1Information processing apparatus, information processing method, information processing system, and computer program
Publication Date: 2022.03.09 KK TOSHIBA
  • EP3965049A1 patent drawingFigure 1
  • EP3965049A1 patent drawingFigure 2
  • EP3965049A1 patent drawingFigure 3(A)~3(B)

AI summary

According to one approach, an information processing apparatus comprises: a first model generator configured to, generate a plurality of first prediction models for an objective variable based on data including an explanatory variable and the objective variable, and a plurality of model generation conditions; and a second model generator configured to weight the plurality of first prediction models based on differences between the objective variable and predicted values of the plurality of first prediction models calculated based on the explanatory variable, and to generate a second prediction model for the objective variable based on the weighted plurality of first prediction models.