CN-121998379-A - New energy automobile charging scheduling method and system based on deep learning

CN121998379ACN 121998379 ACN121998379 ACN 121998379ACN-121998379-A

Abstract

The invention discloses a new energy automobile charging scheduling method and system based on deep learning, which are used for obtaining multi-source data by preprocessing data, establishing a deep learning model by combining a self-attention mechanism of a transducer with a GNN graph neural network, processing a time sequence, generating a time-space joint characteristic by GNN modeling the topological relation between the geographic distribution of charging piles and the load of a power grid, dynamically adjusting the weight of the time-space joint characteristic according to a real-time environment by introducing element learning to obtain a charging demand prediction index, solving an optimal charging pile distribution scheme by MIP mixed integer programming according to the charging demand prediction index, and generating a charging scheduling scheme by adopting a greedy algorithm. The total charging cost is reduced, the average waiting time of users is shortened, and the power grid load variance is reduced.

Inventors

PENG YONG
LIU XIAOZHUANG
WANG ZHE
YUAN JUNGANG
TANG PENG
YAO MAI
LI XIULI

Assignees

贵州万家灯火电气智造有限公司

Dates

Publication Date: 20260508
Application Date: 20260408

Claims (10)

1. The new energy automobile charging scheduling method based on deep learning is characterized by comprising the following steps of: acquiring a user historical charging record, a vehicle battery health state, real-time position information, power grid load data and meteorological data, and preprocessing the data to obtain multi-source data; combining a self-attention mechanism of a transducer with a GNN graph neural network to establish a deep learning model, processing a time sequence by utilizing the transducer, and outputting a space-time joint characteristic by using a topological relation between geographic distribution of a GNN modeling charging pile and a power grid load; introducing meta learning to dynamically adjust the weight of the space-time joint characteristics according to a real-time environment to obtain a charging demand prediction index; and solving an optimal charging pile distribution scheme through MIP mixed integer programming according to the charging demand prediction index, and generating a charging scheduling scheme by adopting a greedy algorithm.
2. The deep learning-based new energy automobile charging scheduling method of claim 1, wherein the obtaining the user history charging record, the vehicle battery health status, the real-time location information, the power grid load data and the meteorological data, preprocessing the data to obtain the multi-source data comprises: acquiring a user historical charging record, a vehicle battery health state, real-time position information, power grid load data and meteorological data; filling time series data by a linear interpolation method, filling classification data by a mode filling method, identifying abnormal values in power grid load and charging quantity by a 3 sigma criterion, and removing invalid data by combining business logic; Uniformly mapping battery temperature, power grid load and position coordinate data to a [0,1] interval by using a Min-Max standardization method; And (3) carrying out time synchronization on the data by taking the time stamp as a reference, uniformly adjusting the time granularity to be 1 minute, matching the real-time position of the vehicle with the coordinates of the charging station, calculating the linear distance between the vehicle and each charging pile and the estimated running time, and obtaining the multi-source data.
3. The deep learning-based new energy automobile charging scheduling method of claim 1, wherein the combining the self-attention mechanism of a transducer with a GNN graph neural network to build a deep learning model, processing a time sequence by using the transducer, GNN modeling the topological relation between the geographic distribution of charging piles and the power grid load, and outputting the time-space joint characteristics comprises: splitting the standardized time sequence data according to variable dimensions, mapping each variable into 64-dimensional vectors through an embedding layer of a transducer, and adding position coding information; constructing a 3-layer self-attention module, setting 8 attention heads on each layer, capturing the law of the change of the charging demand along with time by calculating the attention weights among different time steps, and mining the association among multiple variables; The self-attention output is processed through a two-layer fully-connected Feed Forward network, and the time feature vector is output through residual connection and layer normalization operation.
4. The deep learning-based new energy automobile charging scheduling method of claim 1, wherein the combining the self-attention mechanism of a transducer with a GNN graph neural network to build a deep learning model, processing a time sequence by using the transducer, GNN modeling the topological relation between the geographic distribution of charging piles and the power grid load, and outputting the time-space joint characteristics comprises: Constructing a heterogram of vehicle-charging pile-power grid nodes, wherein the nodes comprise vehicle nodes, charging pile nodes and power grid nodes; Adopting a 2-layer GCN graph rolling network as a core, wherein the first layer aggregates neighbor information of each node, the second layer deeply processes the aggregated characteristics and captures indirect association among multiple nodes; and carrying out global average pooling on the node characteristics output by the GCN, and compressing the characteristics of the iso-composition into spatial characteristic vectors.
5. The deep learning-based new energy automobile charging scheduling method of claim 1, wherein the combining the self-attention mechanism of a transducer with a GNN graph neural network to build a deep learning model, processing a time sequence by using the transducer, GNN modeling the topological relation between the geographic distribution of charging piles and the power grid load, and outputting the time-space joint characteristics comprises: splicing the time feature vector output by the transducer with the space feature vector output by the GNN to obtain an initial fusion feature; And calculating the association weight between each dimension of the time feature and the space feature through the single-head self-attention module, and outputting the space-time joint feature.
6. The deep learning-based new energy automobile charging scheduling method of claim 1, wherein the introducing element learning dynamically adjusts the weight of the space-time joint feature according to a real-time environment to obtain a charging demand prediction index, and comprises the following steps: Extracting environmental characteristics from the multi-source data, wherein the environmental characteristics comprise the current meteorological type, the period type and the real-time load level of the power grid, and obtaining an environmental characteristic vector; The meta learning is matched with a historical meta task according to the environmental feature vector, the adjustment of the weight of each dimension of the time-space joint feature is based on the parameter updating rule of the internal circulation training, the feature strongly related to the current scene is given high weight, and the weight of the weakly related feature is reduced.
7. The deep learning-based new energy automobile charging scheduling method of claim 1, wherein the solving the optimal charging pile allocation scheme according to the charging demand prediction index through MIP mixed integer programming and generating the charging scheduling scheme by adopting a greedy algorithm comprises: Overlapping the charging demand prediction index from high to low and the time for reaching the charging pile from the morning to evening to form a greedy ordering criterion; And collecting all vehicle information of each charging pile, sequencing the vehicles according to a sequencing criterion to form a service queue of the charging pile, calculating the estimated starting charging time and ending charging time of each vehicle, and generating a charging scheduling scheme.
8. New energy automobile charge scheduling system based on degree of depth study, its characterized in that, new energy automobile charge scheduling system includes following module: The multi-source data acquisition module is used for acquiring historical charging records of users, vehicle battery health states, real-time position information, power grid load data and meteorological data, and preprocessing the data to obtain multi-source data; The combined characteristic generation module is used for combining a self-attention mechanism of a transducer with the GNN graph neural network to establish a deep learning model, processing a time sequence by utilizing the transducer, and outputting a time-space combined characteristic by using the topological relation between the geographic distribution of the GNN modeling charging pile and the power grid load; the charging demand prediction module is used for introducing meta learning to dynamically adjust the weight of the space-time joint characteristics according to the real-time environment so as to obtain a charging demand prediction index; and the scheduling scheme generating module is used for solving an optimal charging pile distribution scheme through MIP mixed integer programming according to the charging demand prediction index, and generating a charging scheduling scheme by adopting a greedy algorithm.
9. The deep learning-based new energy vehicle charging schedule system of claim 8, wherein the charging demand prediction module comprises the following submodules: The extraction sub-module is used for extracting environmental characteristics from the multi-source data, including the current meteorological type, the period type and the real-time load level of the power grid, and obtaining an environmental characteristic vector; And the adjustment sub-module is used for matching the meta-learning with the historical meta-task according to the environmental feature vector, updating rules based on parameters of the internal circulation training, adjusting the weight of each dimension of the time-combined feature, giving high weight to the feature strongly related to the current scene, and reducing the weight to the weakly related feature.
10. The deep learning-based new energy vehicle charging schedule system of claim 8, wherein the charging demand prediction module comprises the following submodules: the sequencing sub-module is used for overlapping the greedy charge from the early to the late into a sequencing criterion according to the charge demand prediction index from the high to the low and the time for reaching the charging pile; The generation sub-module is used for collecting all vehicle information of each charging pile, sequencing the vehicles according to a sequencing criterion, forming a service queue of the charging pile, calculating the estimated starting charging time and ending charging time of each vehicle, and generating a charging scheduling scheme.

Description

New energy automobile charging scheduling method and system based on deep learning Technical Field The invention relates to the technical field of new energy automobile charging scheduling, in particular to a new energy automobile charging scheduling method and system based on deep learning. Background Along with the rapid improvement of the permeability of the new energy automobile, the contradiction between the charging demand and the power grid bearing and charging pile layout is increasingly highlighted. The existing charge scheduling method depends on single charge record or position information at the data layer, ignores key influence factors such as power grid load and battery health state and the like to cause incomplete data support, adopts a single time sequence or space modeling method at the model layer, is difficult to consider time periodicity and space relevance of charge requirements, optimizes targets, focuses on single dimension, and is easy to cause problems of power grid load unbalance, high operation cost and low scheduling efficiency. Disclosure of Invention The invention aims to solve the problems and designs a new energy automobile charging scheduling method and system based on deep learning. The technical scheme for achieving the purpose is that in the new energy automobile charging scheduling method based on deep learning, the new energy automobile charging scheduling method comprises the following steps of: acquiring a user historical charging record, a vehicle battery health state, real-time position information, power grid load data and meteorological data, and preprocessing the data to obtain multi-source data; combining a self-attention mechanism of a transducer with a GNN graph neural network to establish a deep learning model, processing a time sequence by utilizing the transducer, and outputting a space-time joint characteristic by using a topological relation between geographic distribution of a GNN modeling charging pile and a power grid load; introducing meta learning to dynamically adjust the weight of the space-time joint characteristics according to a real-time environment to obtain a charging demand prediction index; and solving an optimal charging pile distribution scheme through MIP mixed integer programming according to the charging demand prediction index, and generating a charging scheduling scheme by adopting a greedy algorithm. Further, in the deep learning-based new energy automobile charging scheduling method, the acquiring the user history charging record, the vehicle battery health state, the real-time position information, the power grid load data and the meteorological data, preprocessing the data to obtain multi-source data comprises the following steps: acquiring a user historical charging record, a vehicle battery health state, real-time position information, power grid load data and meteorological data; filling time series data by a linear interpolation method, filling classification data by a mode filling method, identifying abnormal values in power grid load and charging quantity by a 3 sigma criterion, and removing invalid data by combining business logic; Uniformly mapping battery temperature, power grid load and position coordinate data to a [0,1] interval by using a Min-Max standardization method; And (3) carrying out time synchronization on the data by taking the time stamp as a reference, uniformly adjusting the time granularity to be 1 minute, matching the real-time position of the vehicle with the coordinates of the charging station, calculating the linear distance between the vehicle and each charging pile and the estimated running time, and obtaining the multi-source data. Further, in the method for scheduling charging of a new energy automobile based on deep learning, the combining the self-attention mechanism of the fransformer with the GNN graph neural network to establish a deep learning model, processing the time sequence by using the fransformer, GNN modeling the topological relation between the geographic distribution of charging piles and the load of the power grid, and outputting the time-space joint feature comprises: splitting the standardized time sequence data according to variable dimensions, mapping each variable into 64-dimensional vectors through an embedding layer of a transducer, and adding position coding information; constructing a 3-layer self-attention module, setting 8 attention heads on each layer, capturing the law of the change of the charging demand along with time by calculating the attention weights among different time steps, and mining the association among multiple variables; The self-attention output is processed through a two-layer fully-connected Feed Forward network, and the time feature vector is output through residual connection and layer normalization operation. Further, in the method for scheduling charging of a new energy automobile based on deep learning, the combining the self-attention mechani