Machine Learning Energy Management System for Demand Charge Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Commercial and industrial sites face challenges in managing peak demand charges due to unpredictable energy usage patterns and the limitations of existing forecasting methods, particularly when cloud cover affects photovoltaic cell production, leading to potential battery capacity shortages and increased costs.
Innovation Solution
A method and system utilizing machine learning techniques, such as gradient boosted machine learning and support vector regression, to predict demand charges and photovoltaic cell production, incorporating data on load predictability, load shape, battery capacity, weather, and electricity tariffs, to optimize energy management and reserve battery capacity effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional forecasting methods are used to predict energy demand, then the system is simple to implement, but the accuracy of demand prediction is insufficient leading to battery capacity shortages
Solution Approach 1:
The patent replaces traditional mechanical/statistical forecasting methods with machine learning algorithms (gradient boosted machines, support vector regression) to predict energy demand and photovoltaic production. This substitution enables the system to capture complex non-linear patterns in energy consumption and generation data, significantly improving prediction accuracy while automating the forecasting process through computational models that adapt to varying conditions including cloud cover effects.
2Measurement precision
If machine learning techniques are used to improve forecasting accuracy, then the prediction precision improves, but the computational complexity and data processing requirements increase
Solution Approach 1:
The system performs preliminary data processing and feature engineering before applying machine learning models. Historical energy consumption data, photovoltaic production data, and weather data are pre-processed, aggregated, and transformed into suitable formats. This preliminary action reduces the computational burden during actual forecasting operations and improves model training efficiency.
Solution Approach 2:
The patent uses gradient boosted machines and support vector regression models that create simplified representations (copies) of complex energy system behaviors. These machine learning models learn from historical data and create predictive functions that approximate the relationship between input features (weather, time, historical data) and output predictions (energy demand, PV production), reducing the need for complex real-time calculations.
3Reliability
If battery capacity is increased to handle unexpected demand spikes, then the reliability of energy supply improves, but the cost and space requirements increase
Solution Approach 1:
The system performs preliminary forecasting of energy demand and photovoltaic production to anticipate future energy needs and surplus generation. By predicting demand spikes and PV production variations in advance, the system can optimally charge or discharge battery storage beforehand, ensuring sufficient capacity is available during peak demand periods without requiring excessive battery capacity to be permanently maintained at full charge.
Solution Approach 2:
The patent implements a feedback mechanism where actual energy consumption and PV production data are continuously monitored and compared with forecasted values. This feedback loop allows the system to learn from prediction errors and adjust future forecasts, while also dynamically managing battery charge/discharge operations based on real-time conditions, optimizing the use of available battery capacity.
4Loss of energy
If demand charge management is implemented to reduce peak charges, then the electricity cost decreases, but the system complexity and control requirements increase
Solution Approach 1:
The energy management system operates autonomously using machine learning models to automatically predict energy demand, forecast PV production, and control battery charge/discharge operations. The system makes independent decisions about when to draw from the grid, when to use stored energy, and when to export surplus PV production, eliminating the need for constant manual intervention or complex centralized control while achieving demand charge reduction.
Solution Approach 2:
The patent replaces manual energy management and complex control systems with automated machine learning-based decision-making. Gradient boosted machines and support vector regression models automatically analyze patterns in energy consumption and generation data to optimize energy dispatch strategies, substituting human expertise and complex mechanical control systems with adaptive computational algorithms that continuously optimize for cost reduction.
Data Source
AI summary
Methods, systems, and techniques for determining the confidence that predicted demand charge savings resulting from deploying an energy management system at a site will match actual cost savings. A processor obtains predicted demand charge savings resulting from deploying the energy management system at the site, and also obtains demand-related data describing at least two of load predictability, load shape, and battery capacity of the energy management system. The processor determines a confidence score representing the confidence that the predicted demand charge savings will result in view of the demand-related data, and displays the confidence score on a display for subsequent use by a user.