Predicting Electric Vehicle Charging Station Usage from Historical Data

越来越重要的是，公用事业公司必须准确了解其电网将需要提供多少能源，以支持电动汽车和充电基础设施的快速增长。与此同时，电动汽车用户想知道在哪里给汽车充电最好。Mogile Technologies运营着ChargeHub，这是一个帮助消费者找到移动充电站的平台。为了支持这一产品，他们汇总了充电站使用情况的数据；如果他们可以使用这些数据来预测使用情况，这将极大地改善和扩展他们可以为公用事业公司和消费者提供的服务。该项目的目标是识别最能预测未来充电站使用情况的机器学习模型。我们将探索各种方法，从回归到深度学习；我们将使用各种输入特征，包括那些表征过去充电事件的输入特征(例如，每次充电能量)，以及在时间允许的情况下，预期提高准确性的额外变量(例如，天气条件)。作为领域专家，Mogile将与我们密切合作，以确定构建一组训练数据的适当战略。作为机器学习专家，我们将建立和评估模型，这些模型给定(A)训练数据和(B)过去的使用情况，(C)预测未来七天的使用情况，以及(D)相关的预测置信度。这项工作将以原型API的形式交付，并以交付报告的形式进行记录。

It is increasingly important that utilities have a precise view of how much energy their grid will need to provide to support the rapid growth of electric vehicles and the charging infrastructure. Meanwhile, electric vehicle users want to know where best to charge their vehicles.Mogile Technologies operates ChargeHub, a platform that, amongst other things, helps consumers find charging stations on the go. In support of this product, they aggregate data on charging station usage; if they can use this data to predict usage, it will dramatically improve and expand the services they can provide to both utilities and consumers alike.The objective of this project is to identify machine learning models that best predict future charging station usage. We will explore a variety of approaches, from regression to deep learning; we will work with a variety of input features, including those the characterize past charging events (e.g., per charge energy), and, time permitting, additional variables expected improve accuracy (e.g., weather conditions). As domain experts, Mogile will work with us closely to identify the appropriate strategy for constructing a set of training data. As machine learning experts, we will build and evaluate models that, given (a) training data, and (b) past usage, (c) predicts usage over the next seven days, and (d) associated prediction confidence. This work will delivered in the form of a prototype API, and documented in the form of a deliverable report.