Power Demand Forecasting With Real-Time Error Correction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current power management systems lack reliability in predicting power consumption due to the accumulation of errors when using past data to forecast future consumption, leading to inefficiencies and inaccuracies in demand management.

Innovation Solution

A closed-loop power management system employing a full-time scale prediction method with machine learning models for both short-term and long-term predictions, coupled with real-time error tracking and model adjustment, to improve prediction accuracy and reduce the difference between actual and predicted demand.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If prediction data from previous time points is used to calculate future power consumption data, then the prediction process can be performed, but accumulation of errors occurs leading to reduced reliability

Engineering Contradiction:
Improveprediction reliabilityVSAvoidprediction accuracy
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The patent segments the prediction task into multiple independent prediction models, each trained on specific historical time periods. Instead of using sequential prediction where errors accumulate, the system divides the prediction process into separate segments (different time periods) that can be independently calculated and compared with actual data, preventing error propagation across the entire prediction chain.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent changes the parameter of time period segmentation by training separate prediction models on different historical time periods. This parameter change allows the system to capture temporal variations in power consumption patterns while maintaining prediction independence, thereby improving both reliability and accuracy without suffering from error accumulation.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If multiple prediction models are used for different time periods, then prediction accuracy improves, but system complexity increases

Engineering Contradiction:
Improveprediction accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system segments the prediction task into multiple models trained on different time periods, which improves accuracy by capturing temporal patterns. The complexity is managed by organizing these segmented models in a structured manner where each model handles a specific time period, making the overall system complexity tractable and manageable.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The prediction system serves multiple functions: it predicts power consumption for different time periods, evaluates prediction accuracy, identifies high-quality power users, and optimizes power supply plans. This multi-functionality justifies the use of multiple prediction models, as the system gains comprehensive capabilities that outweigh the increased complexity.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Measurement precision

If real-time error tracking is implemented, then prediction accuracy improves, but computational requirements and processing time increase

Engineering Contradiction:
Improveprediction accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary actions by pre-training multiple prediction models on historical data from different time periods before actual prediction is needed. This preliminary preparation allows the models to be ready for rapid inference, reducing the processing time required during real-time operation while maintaining high prediction accuracy through error tracking.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3961841B1Power management method and apparatus, computing device, medium, and product
Publication Date: 2024.10.23 SIEMENS (CHINA) CO LTD
  • EP3961841B1 patent drawingFigure 1
  • EP3961841B1 patent drawingFigure 2
  • EP3961841B1 patent drawingFigure 3

AI summary

The present disclosure relates to a power management method and apparatus, a computing device, a medium, and a product. The power management method comprises: a monitoring step: in real time, acquiring and storing power consumption data of a user; a prediction step: on the basis of the acquired power consumption data, using a pre-stored first prediction model to predict a first predicted power consumption of a first time period, and using a pre-stored second prediction model to predict a second predicted power consumption of each second time period included in the first time period; an error calculation step: on the basis of the acquired real-time power consumption data, from the first second time period to the last second time period included in the first time period, sequentially executing the following processing on each second time period: calculating a first error between the actual power consumption of the second time period and the second predicted power consumption of the second time period, and calculating a second error between the total actual power consumption from the first second time period to the current second time period and the sum of the each of the second predicted power consumptions from the first second time period to the current second time period; and an adjustment step: when the first error is greater than a first predetermined threshold and/or the second error is greater than a second predetermined threshold, adjusting the power supply plan or the power demand of the user.