Forecasting Substation Load With Sparse Instrumentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for monitoring electrical substations require full instrumentation, which is costly and inefficient, as they need to instrument all substations for accurate load forecasting, leading to high deployment costs and potential overspending.
Innovation Solution
The use of unsupervised machine learning to identify representative instrumented substations that can forecast behavior of noninstrumented ones, combined with external data sources like temperature and weather sensors, and supervised machine learning to generate predictive models, allowing for sparse instrumentation deployment strategies that reduce costs while maintaining forecasting accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If full instrumentation is deployed at all substations, then forecasting accuracy is improved, but deployment cost increases significantly
Solution Approach 1:
The patent implements partial instrumentation by deploying monitoring devices at only a subset of substations rather than all substations. The system uses machine learning models to forecast load at non-instrumented substations based on data from instrumented ones, achieving acceptable forecasting accuracy with reduced deployment costs.
Solution Approach 2:
The patent creates virtual copies of instrumented substations through machine learning models. By training models on data from instrumented substations, the system generates predictive representations that estimate load behavior at non-instrumented substations, effectively copying the monitoring capability without physical instrumentation.
2Quantity of substance
If sparse instrumentation is used, then deployment cost is reduced, but forecasting accuracy deteriorates
Solution Approach 1:
The patent introduces machine learning models as intermediaries between instrumented and non-instrumented substations. These models act as mediators that transfer knowledge from instrumented substations to predict behavior at non-instrumented ones, enabling accurate forecasting without direct measurement at every location.
Solution Approach 2:
The patent performs preliminary training of machine learning models using historical data from instrumented substations before deployment. This preliminary action prepares the models to accurately forecast load at non-instrumented substations, ensuring forecasting accuracy is maintained even with sparse instrumentation.
3Quantity of substance
If machine learning models are trained on limited data from sparse instrumentation, then deployment cost is reduced, but model reliability deteriorates
Solution Approach 1:
The patent merges data from multiple instrumented substations to train machine learning models that can generalize to non-instrumented substations. By combining information from multiple sources and using ensemble methods, the system improves model reliability even when each individual substation has limited data.
Data Source
AI summary
Systems and a method for forecasting data at noninstrumented substations from data collected at instrumented substations is provided. An example method includes determining a cluster id for a noninstrumented substation, creating a model from data for instrumented substations having the cluster id, and forecasting the data for the noninstrumented station from the model.