EV Charging Time Forecasting With Clustered ML Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing charging time forecasting methods for electric vehicles are not reliable due to factors like varying vehicle types, charging station configurations, and external conditions, leading to inconsistent and often longer-than-expected charging times.
Innovation Solution
A method using machine learning to generate charging time forecasting models that consider not only vehicle-specific parameters but also external factors such as operating conditions of the vehicle and charging station, weather, and battery status, with a self-learning system that improves accuracy over time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If charging power is distributed between multiple charging stations at a congested charging site, then more vehicles can be served simultaneously, but charging time increases because vehicles do not receive maximal charging power
Solution Approach 1:
The system performs preliminary actions by forecasting charging times before vehicles arrive at charging stations. The machine learning model predicts charging duration based on historical data, vehicle characteristics, and station status, allowing users to plan routes and charging stops in advance, thereby reducing waiting time and improving overall system efficiency
2Device complexity
If machine learning models use only basic vehicle parameters for forecasting, then the system is simpler to implement, but forecasting reliability is insufficient due to varying external conditions
Solution Approach 1:
The system applies parameter changes by continuously expanding the feature set used in machine learning models. Beyond basic vehicle parameters, the system incorporates charging station characteristics, historical charging data, weather conditions, and vehicle operating status. The model dynamically adjusts parameters based on data availability and relevance, improving forecasting reliability while managing complexity through selective parameter inclusion
Solution Approach 2:
The system implements feedback mechanisms where actual charging outcomes are fed back into the machine learning model to continuously improve predictions. The model learns from discrepancies between predicted and actual charging times, adjusting its parameters and weightings to enhance accuracy over time. This closed-loop approach allows the system to adapt to changing conditions and improve reliability without requiring complete system redesign
Data Source
AI summary
The present invention relates to a system and a method for forecasting charging time of electric vehicles. Charging event data is filtered to discard at least one piece of the charging event data or disable use of at least one piece of the charging event data. The remaining charging event data is processed for generating a data collection comprising sample data. A data cluster representing sample data associated with a selected electric vehicle cluster is obtained and a charging time forecasting model is generated on basis of the obtained data cluster using machine learning. A plurality of charging time forecasting models are trained based on a first portion of the data cluster. The best performing charging time forecasting model is selected on basis of a second portion of the data cluster, and the selected charging time forecasting model is tested based on a third portion of the data cluster. The charging time forecasting model is applied on a charging event of an electric vehicle comprised in the selected electric vehicle cluster.