Data Analysis System Dynamic Clustering for Energy Demand

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing energy demand analysis systems struggle to classify time series data and consumption patterns accurately, leading to potential deficiencies or excesses in power generation and procurement, especially for new consumers without historical data and those with unacquired patterns.

Innovation Solution

A data analyzing system that dynamically determines the optimal cluster number for load data classification based on intra-cluster relevance and inter-cluster separation, using clustering algorithms like k-means, and generates diagnostic decision trees to classify new consumers, ensuring accurate grouping and energy demand forecasting.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If the cluster number is set manually for classifying time series data, then the classification process is simple, but the classification accuracy does not match the actual situation

Engineering Contradiction:
Improveclassification process simplicityVSAvoidclassification accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The system automatically determines the optimal cluster number by evaluating intra-cluster relevance and inter-cluster separation metrics, eliminating the need for manual cluster number specification. The classification system serves itself by selecting appropriate parameters based on data characteristics, thereby resolving the contradiction between operational simplicity and classification accuracy.

Inventive Principle:
Principle #25Self-service

2Device complexity

If groups of consumption patterns are set in advance, then the system structure is simple, but the system cannot adapt to new consumers without historical data

Engineering Contradiction:
Improvesystem structureVSAvoidadaptability to new consumers
Core Design Contradiction:
Device complexityVSAdaptability or versatility

Solution Approach 1:

The system dynamically determines cluster assignments and consumption pattern groupings based on available data characteristics rather than relying on pre-defined static groups. For new consumers without historical data, the system adapts by using alternative features and dynamically adjusting cluster assignments, thereby achieving adaptability while maintaining reasonable system structure.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system performs preliminary classification using available features (even if incomplete) to assign new consumers to appropriate clusters before full data becomes available. This preliminary action enables the system to handle new consumers immediately rather than waiting for complete historical data, improving adaptability while keeping the overall system structure manageable.

Inventive Principle:
Principle #10Preliminary action

3Ease of manufacture

If manual cluster number setting is used, then the system is easy to implement, but power generation adjustment does not match actual demand

Engineering Contradiction:
Improvesystem implementation easeVSAvoidpower generation adjustment accuracy
Core Design Contradiction:
Ease of manufactureVSReliability

Solution Approach 1:

The system incorporates feedback mechanisms that evaluate classification quality metrics (intra-cluster relevance and inter-cluster separation) to automatically adjust the optimal cluster number. This feedback loop ensures that power generation adjustments are based on accurate, data-driven cluster assignments rather than fixed manual settings, thereby improving reliability while maintaining implementation feasibility through automated optimization.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP3115959B1Data analysis system and method
Publication Date: 2020.06.03 HITACHI LTD
  • EP3115959B1 patent drawingFigure 1
  • EP3115959B1 patent drawingFigure 2
  • EP3115959B1 patent drawingFigure 3

AI summary

Proposed is a data analyzing system and method capable of performing highly reliable analytical processing which matches the actual situation. In a data analyzing device and method in which load data is classified into a plurality of clusters in consumer units based on the load data representing the power usage of each consumer for each unit time and the attribute information of each consumer, a diagnostic decision tree is generated for classifying the consumer into one of the clusters based on the attribute information of that consumer.