A charging demand prediction system for new energy vehicle charging stations based on big data

Abstract

The present invention proposes a charging demand prediction system for new energy vehicle charging stations based on big data, which relates to the technical field of charging demand prediction. The specific implementation steps include: first, collecting vehicle images and identifying them through high-definition cameras to obtain vehicle flow data, and at the same time obtaining charging pile distribution, geographical distribution, historical charging data and meteorological data; second, cleaning and processing the acquired data, and constructing a charging station spatial relationship diagram; then, by analyzing the similarity of historical data sequences, determining time series key data; finally, based on the charging station spatial relationship diagram and time series key data, constructing a charging quantity prediction model, thereby predicting future charging quantity and determining whether to update the model. The system not only improves the accuracy of charging demand prediction, but also optimizes the resource configuration of charging stations. It can monitor and dynamically adjust the charging strategy of charging stations in real time, thereby improving the utilization efficiency of charging facilities.

Description

Technical Field

[0001] The present invention relates to the technical field of charging capacity demand prediction, and in particular to a charging capacity demand prediction system for a new energy vehicle charging station based on big data. Background Art

[0002] New energy vehicles have the advantages of high energy efficiency, low operating costs and easy maintenance, making new energy vehicles gradually replace fuel vehicles and become the mainstream means of transportation in modern society. Charging stations are the main place for new energy vehicle users to charge, and their number, geographical distribution and performance of charging piles directly affect the user's charging experience. However, current charging stations have problems such as uneven distribution and insufficient or excessive number of charging piles. Especially in highway service areas and popular tourist attractions, charging equipment is often in short supply or idle, resulting in low operating efficiency of charging stations. Therefore, through charging demand forecasting, the future charging demand of the corresponding charging station can be obtained, so as to reasonably allocate charging pile resources and improve the operating efficiency of the charging station.

[0003] At present, the existing charging demand forecast is mainly based on the historical charging data of the charging station, the distribution of residents, schools and enterprises around the charging station, and the charging demand of urban public charging stations. It cannot be directly applied to the charging demand forecast of new energy vehicle charging stations in highway service areas. In addition, the previous forecasting method ignores the sudden changes in charging demand caused by factors such as weather, holidays and tourist peak seasons, making it difficult to accurately capture changes in user behavior, further affecting the rational allocation of charging resources.

[0004] In order to solve the above problems, a charging demand prediction system for new energy vehicle charging stations based on big data is proposed. Summary of the invention

[0005] The purpose of the present invention is to provide a new energy vehicle charging station charging demand prediction system based on big data, which is used to predict the future charging demand of the charging station. In order to solve the problems existing in the prior art, the present invention first collects and identifies vehicle images through high-definition cameras to obtain vehicle flow data, and at the same time obtains charging pile distribution, geographical distribution, historical charging data and meteorological data; secondly, the acquired data is cleaned and processed, and a charging station spatial relationship diagram is constructed; then, by analyzing the similarity of historical data sequences, the time series key data is determined; finally, based on the charging station spatial relationship diagram and time series key data, a charging capacity prediction model is constructed to predict future charging capacity and determine whether to update the model. The system not only improves the accuracy of charging demand prediction, but also optimizes the resource configuration of the charging station. It can monitor and dynamically adjust the charging strategy of the charging station in real time, thereby improving the utilization efficiency of charging facilities.

[0006] A charging demand prediction system for new energy vehicle charging stations based on big data, the specific implementation steps include:

[0007] A data acquisition module acquires charging pile distribution data, geographical distribution data, historical charging data, meteorological data and vehicle images of target charging stations in high-speed service areas; and obtains vehicle flow data by combining image recognition technology with the vehicle images;

[0008] The data processing module cleans the charging data, meteorological data and vehicle flow data to obtain historical charging data, historical meteorological data and historical vehicle flow data; obtains the spatial relationship between the target charging station and the adjacent charging stations and the adjacent tourist destinations according to the geographical distribution data; obtains the charging pile data in the adjacent charging station, the rating and the tourist ratio of the adjacent tourist destination, and constructs a charging station spatial relationship diagram in combination with the charging pile distribution data;

[0009] The data screening module obtains the historical data after data cleaning, calculates the first attribute coefficient and the second attribute coefficient between the historical data sequences with different attributes, and generates a historical data map; removes redundant attributes in the historical data through threshold screening to obtain time series key data;

[0010] The charging capacity prediction module constructs and trains a charging capacity prediction model based on the charging station spatial relationship diagram and the key time series data of historical time; obtains the correlation between neighboring charging stations, neighboring tourist destinations and date types and charging capacity through the model, and predicts future charging capacity; monitors the actual charging capacity of the target charging station in real time, and compares it with the prediction result to determine whether to update the model; if necessary, fine-tunes the model parameters through an online update strategy.

[0011] Preferably, the process of obtaining the vehicle flow data by combining the image recognition technology and the vehicle image includes:

[0012] According to the vehicle images, the vehicle images are screened by image recognition technology, and images of different vehicles are retained to obtain the total number of vehicles per day; in combination with the image segmentation algorithm, the license plate area of ​​the retained image is located and segmented to obtain the corresponding license plate of each vehicle image; based on the color analysis of the license plate image, the daily number of new energy vehicles is determined to obtain the proportion of new energy vehicles; wherein the traffic flow data includes the total number of vehicles and the proportion of new energy vehicles; the vehicle images are collected by high-definition cameras installed at the entrances and exits of the highway service area.

[0013] Preferably, the specific implementation process of constructing the charging station spatial relationship diagram includes:

[0014] Determine the area around the target charging station based on the geographic distribution data The coordinates of the adjacent charging stations and the adjacent tourist destinations in the area; obtaining the first mileage distance between the target charging station and the adjacent charging station, the second mileage distance between the adjacent tourist destination and the charging station, and the charging pile data in the adjacent charging station, and establishing a spatial graph of the target charging station; the spatial graph includes: spatial nodes and edges; the spatial nodes are the target charging station, the adjacent charging station, and the adjacent tourist destination; combining the charging pile distribution data, the first mileage distance, the second mileage distance, the charging pile data in the station, the score of the adjacent tourist destination and the proportion of tourists, calculate the edge weights between different spatial nodes; if the edge weight between the adjacent charging station node and the adjacent tourist destination node is less than or equal to the predetermined spatial weight, delete the corresponding edge in the spatial graph to obtain the charging station spatial relationship graph.

[0015] Preferably, the implementation process of the data screening module includes:

[0016] According to the historical charging data, the historical meteorological data and the historical vehicle flow data, a first historical data sequence set, a second historical data sequence set and a third historical data sequence set are respectively obtained, wherein the historical data sequence set contains the historical data sequences of different attributes; the similarity between the historical data sequences of different attributes in the same historical data sequence set is obtained to obtain the first attribute coefficient; the similarity between the historical data sequences in different historical data sequence sets is obtained to obtain the second attribute coefficient; the historical data sequences of different attributes, the first attribute coefficient and the second attribute coefficient are combined to generate the historical data map; the historical data map includes: nodes and edges; the nodes are the historical data sequences; the edges represent the relationship between different nodes, and the edge weights are the first attribute coefficient and the second attribute coefficient.

[0017] Preferably, the implementation process of the data screening module also includes:

[0018] The historical data map is screened according to the first attribute coefficient and the second attribute coefficient; if the first attribute coefficient is greater than a predetermined first weight threshold, the sum of the first attribute coefficients of two corresponding nodes is calculated respectively, and the node with the largest sum of the first attribute coefficients is deleted; if the first attribute coefficient is less than or equal to the predetermined first weight threshold, the second attribute coefficient of the corresponding node is determined; if the second attribute coefficient is greater than the predetermined second weight threshold, the corresponding node is deleted; the historical data sequences corresponding to all retained nodes are spliced ​​to obtain time series key data.

[0019] Preferably, the specific implementation process of the charging capacity prediction module includes:

[0020] In the training phase, the time series key data of the historical time is divided to obtain training samples and true values; the training samples and the charging station spatial relationship diagram are input into the charging capacity prediction model to obtain the predicted charging capacity; the model parameters are optimized based on the true value and the predicted charging capacity to obtain the trained charging capacity prediction model;

[0021] In the prediction stage, the key time series data within a preset time period before the day to be predicted is obtained, and the key time series data is input into the trained charging capacity prediction model to obtain the future charging capacity of the day to be predicted;

[0022] In the model updating stage, the charging piles of the target charging station are scheduled according to the future charging amount and the charging pile distribution data; the actual charging amount of the target charging station is monitored in real time and compared with the future charging amount to determine whether to update the charging amount prediction model.

[0023] Preferably, the charging capacity prediction model includes:

[0024] The spatial relationship diagram of the charging station and the key time series data are input into the feature extraction layer to obtain the spatial feature sequence and the time series feature sequence; through the spatiotemporal fusion layer, the spatiotemporal correlation of the spatial feature sequence and the time series feature sequence is captured, and the potential trends affecting the charging amount are extracted; combined with the fully connected layer and the prediction head, the future charging amount of the target charging station is obtained.

[0025] Preferably, the specific implementation process of inputting the charging station spatial relationship diagram and the time sequence key data into the feature extraction layer to obtain the spatial feature sequence and the time sequence feature sequence is as follows:

[0026] The spatial features of the target charging station in the charging station spatial relationship diagram are extracted through a graph convolutional network and converted into an adjacency matrix to obtain the spatial feature sequence; the key timing data is input into a convolutional neural network to capture the charging amount change characteristics, user charging behavior characteristics and weather change characteristics to obtain the timing feature sequence.

[0027] Preferably, the space-time fusion layer includes:

[0028] The temporal feature sequence is input into a bidirectional long short-term memory network to capture the correlation between different date types and the dynamic changes in charging capacity, and obtain the first reconstruction feature; through a dense convolutional neural network, the influence of different attribute data features on the charging capacity is extracted to obtain the second reconstruction feature; the spatial feature sequence and the second reconstruction feature are input into a cross-attention network to obtain the dependency relationship between the charging stations and tourist attractions around the target charging station and the changes in charging capacity, and obtain the spatiotemporal fusion feature.

[0029] Preferably, the implementation process of the model updating stage includes:

[0030] The actual charging capacity of the target charging station is monitored in real time and compared with the future charging capacity to obtain a charging capacity error; if the charging capacity error is greater than a predetermined error threshold, the trained charging capacity prediction model is fine-tuned according to the actual charging capacity and the key timing data within the preset time, combined with the online update strategy. The specific formula is expressed as follows:

[0031] ;

[0032] ;

[0033] in, represents the model parameters after fine-tuning; represents the parameters of the original model; represents the learning rate; Express Find the gradient; A loss function representing the charge capacity prediction model; Indicates The predicted value of samples; No. The true value of samples; Represents the mean of all sample prediction values; Indicates the number of samples; represents the adjustment parameter, The value of Within the range and .

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] 1. The present invention combines the distribution of charging piles, geographic information, historical charging data and meteorological data, which not only improves the accuracy of charging demand prediction, but also optimizes the configuration of charging station resources. As the main resting area for mobile vehicles, the charging demand of highway service areas is significantly affected by factors such as weather changes, holiday traffic peaks and tourist seasons. By constructing a charging station spatial relationship diagram, the spatial dependency relationship between the target charging station in the highway service area and the adjacent charging stations and tourist destinations can be accurately described, so that the system can more flexibly capture the charging demand fluctuations caused by these influencing factors, thereby improving the prediction accuracy.

[0036] 2. The present invention uses a charging capacity prediction model to deeply analyze the spatial dependency between charging stations in high-speed service areas and surrounding charging stations and tourist attractions, accurately captures the subtle differences in user charging behavior over geography and time, and realizes the mining of charging demand change trends. The prediction model can quickly capture the mutation characteristics of charging demand during special periods (such as tourist peak seasons), thereby improving the accuracy of charging capacity prediction.

[0037] 3. The model update mechanism proposed in the present invention dynamically adjusts the model parameters according to the error between the real-time charging data and the predicted amount. Through the monitoring and feedback mechanism of the charging error, the model parameters can be adjusted in time to ensure that the prediction model continues to adapt to the fluctuation of charging demand, especially in special circumstances such as holiday peaks and sudden weather changes, to ensure the reliability and stability of the prediction results, so as to adapt to the charging prediction in complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A flowchart of the steps for implementing a new energy vehicle charging station charging demand prediction system based on big data proposed in an embodiment of the present invention;

[0039] Figure 2 A flowchart of a charging demand prediction system for new energy vehicle charging stations based on big data proposed in an embodiment of the present invention;

[0040] Figure 3 A schematic diagram of a charging station space relationship diagram proposed in an embodiment of the present invention;

[0041] Figure 4 This is a structural diagram of the charging capacity prediction model proposed in the embodiment of the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] With the popularity of new energy vehicles, the construction of charging stations in highway service areas has become increasingly important. These charging stations not only provide convenient charging services for new energy vehicles on long-distance trips, but also effectively alleviate the problem of insufficient charging facilities in cities. However, due to the large changes in the charging demand for new energy vehicles, service area charging stations face huge challenges in the number of charging piles and the layout of charging facilities. Therefore, by analyzing historical charging data and accurately predicting the charging demand of charging stations, it can help charging stations to reasonably allocate charging resources and optimize the number and distribution of charging piles, thereby avoiding resource waste and users waiting in line.

[0044] The present invention proposes a new energy vehicle charging station charging demand prediction system based on big data to predict the future charging demand of a target charging station to improve the operating efficiency of the charging station. In order to illustrate the effectiveness of the system proposed by the present invention, a specific description will be given in combination with the accompanying drawings of this embodiment and the following two embodiments.

[0045] Embodiment 1

[0046] The present application embodiment discloses a new energy vehicle charging station charging demand prediction system based on big data to predict the future charging demand for the charging station A in the highway service area. Figure 1 The specific implementation steps of the system proposed in the present invention include: S1. Acquire the spatial distribution data, charging data and vehicle flow data of the target charging station in the high-speed service area; S2. Clean the acquired data and construct a spatial relationship diagram of the charging station; S3. Filter the historical data after data cleaning through the historical data map to obtain time series key data; S4. Construct and train a charging capacity prediction model based on the charging station spatial relationship diagram and time series key data; S5. Input the time series key data within a preset time period into the charging capacity prediction model to obtain a prediction result; S6. Monitor the actual charging capacity in real time and determine whether to update the charging capacity prediction model.

[0047] See also Figure 2 The charging demand prediction system includes: a data acquisition module, a data processing module, a data screening module and a charging demand prediction module; wherein the data acquisition module acquires the charging pile distribution data, geographical distribution data, charging data of historical time and meteorological data of the target charging station in the expressway service area; according to the high-definition camera set at the entrance and exit of the expressway service area, the vehicle images of historical time are collected; the vehicle images are identified by image recognition technology to obtain the vehicle flow data, corresponding to the above-mentioned step S1; the charging pile distribution data includes the number and type of charging piles in the target charging station; the charging data includes date, date type, charging pile type, charging time, user data and charging amount; the vehicle flow data includes the total number of vehicles per day and the proportion of new energy vehicles.

[0048] Furthermore, based on the collected historical vehicle images, image recognition technology is used to detect and identify them, so that the vehicle images are screened and only images of different vehicles are retained to obtain the total number of vehicles per day. Combined with the image segmentation algorithm, the license plate area of ​​the retained image is located and segmented to obtain the corresponding license plate for each vehicle image. Based on the color analysis of the license plate image, the daily number of new energy vehicles is determined to obtain the proportion of new energy vehicles.

[0049] The embodiment of the present application combines image recognition technology and image segmentation algorithm to efficiently and accurately process the collected vehicle images. By screening the images of different vehicles and extracting the license plate information, not only the total number of vehicles per day is obtained, but also the accurate identification and quantity statistics of new energy vehicles are realized. This process improves the accuracy and real-time performance of vehicle flow data, thereby providing reliable basic data for charging demand prediction, helping to understand changes in user behavior, and effectively improving the efficiency of charging facilities.

[0050] In order to clearly understand the historical charging data collected by charging station A in the highway service area, some charging data of charging station A in the past month are given as an example, see Table 1; among them, the historical charging data includes date, date type, charging pile type, charging time, charging amount and user charging data; user charging data includes the amount before charging and the amount after charging.

[0051] Table 1. Some charging data of charging station A in the past month

[0052]

[0053] Furthermore, the data processing module cleans the charging data, meteorological data and vehicle flow data of the historical time to obtain historical charging data, historical meteorological data and historical vehicle flow data; constructs a charging station spatial relationship diagram according to the geographical distribution data and the charging pile distribution data, corresponding to the above-mentioned step S2; cleans the charging data, meteorological data and vehicle flow data of the historical time, and the specific implementation process includes:

[0054] Missing value processing: checking the missing values ​​of the historical data, and performing sample deletion for the missing values ​​of the historical charging data and the historical vehicle flow data; and filling the missing values ​​of the historical meteorological data by linear interpolation of the data at adjacent time points;

[0055] Outlier processing: using machine learning methods such as support vector machines or isolation forests to identify outliers in historical data and remove samples corresponding to outliers;

[0056] Time alignment: align all historical data in chronological order to ensure the consistency of data time;

[0057] Data normalization: convert date type and charging pile type into digital form through dictionary mapping; combine the normalization formula to scale the data to Within the range.

[0058] Furthermore, the specific implementation process of the data processing module constructing the charging station spatial relationship diagram according to the geographical distribution data and the charging pile distribution data includes:

[0059] Determine the area around the target charging station based on the geographic distribution data The coordinates of the neighboring charging stations and neighboring tourist attractions in the area; obtaining the first mileage distance between the target charging station and the neighboring charging station, the second mileage distance between the neighboring tourist attraction and the charging station, and the charging pile data in the neighboring charging station; taking the target charging station, the neighboring charging station, and the neighboring tourist attraction as nodes, combining the charging pile distribution data, the first mileage distance, the second mileage distance, the charging pile data in the station, the score of the neighboring tourist attraction, and the proportion of tourists, calculate the edge weights between different nodes, and the specific formula is expressed as follows:

[0060] ;

[0061] ;

[0062] in, Indicates the target charging station node and the The edge weights between the adjacent charging station nodes, is a positive integer; Indicates The charging station node and The edge weights between the neighboring tourist destination nodes, the target charging station represents the Charging station nodes; Indicates the number of charging piles in the charging pile distribution data; Indicates The charging pile data of each of the adjacent charging stations; Indicates that the target charging station is a first mileage distance between the adjacent charging stations; The average of all first mileage distances; represents a natural constant; Indicates The ratings of the nearby tourist destinations and ; Indicates The proportion of tourists to the neighboring tourist destinations is normalized according to the average number of tourists per year of all the neighboring tourist destinations; Indicates The charging station node and The second mileage distance between the adjacent tourist destination nodes; represents the average value of the second mileage distance;

[0063] According to the node and edge weights, a spatial graph of the target charging station is established, see Figure 3 If the edge weight between the neighboring charging station node and the neighboring tourist destination node is less than or equal to the predetermined spatial weight, the corresponding edge in the spatial graph is deleted to obtain the charging station spatial relationship graph.

[0064] The embodiment of the present application can effectively integrate the geographical distribution information of the target charging station and the characteristics of the surrounding environment by constructing a charging station spatial relationship diagram. The spatial relationship diagram can intuitively display the spatial relationship between each charging station and the surrounding tourist destinations, which helps to analyze the spatiotemporal changes in charging demand and improve the accuracy of subsequent charging volume prediction.

[0065] Furthermore, the data screening module obtains the similarity between historical data sequences of different attributes according to the historical charging data, the historical meteorological data and the historical vehicle flow data, and generates a historical data map; screens the historical data sequences by comparing the similarities, and determines that the time series key data corresponds to the above step S3; the specific implementation process includes:

[0066] According to the historical charging data, the historical meteorological data and the historical vehicle flow data, a first historical data sequence set, a second historical data sequence set and a third historical data sequence set are respectively obtained, and the historical data sequence set contains the historical data sequences of different attributes; through the Pearson correlation coefficient, the similarity between the historical data sequences of different attributes in the same historical data sequence set is obtained to obtain the first attribute coefficient, which is specifically expressed as follows:

[0067] ;

[0068] in, Indicates The first The attributes and The first attribute coefficient between the attributes; Indicates The first and The covariance of the historical data series of the attributes; Indicates The first The standard deviation of the historical data series of the attribute; Indicates The first The standard deviation of the historical data series of the attribute;

[0069] The similarity between the historical data sequences in different historical data sequence sets is obtained to obtain the second attribute coefficient. The specific calculation formula is:

[0070] ;

[0071] ;

[0072] in, Indicates The first attributes and the The first The second attribute coefficient between the attributes; represents the mutual information function; Represents the maximum value of the mutual information between all historical data sequences with different attributes; Indicates The first The attributes correspond to the values ​​of the historical data sequence; Indicates The first The attributes correspond to the values ​​of the historical data sequence; Representation attributes and The joint probability distribution of Representation attributes and The marginal probability distribution of ; Represents a logarithmic function with a natural constant as base;

[0073] The historical data graph is generated by combining the historical data sequence of the different attributes, the first attribute coefficient and the second attribute coefficient; the historical data graph includes: nodes and edges; the nodes are the historical data sequences; the edges represent the relationship between different nodes, and the edge weights are the first attribute coefficient and the second attribute coefficient.

[0074] The embodiment of the present application can effectively identify and analyze the similarities between historical data sequences of different attributes and generate a historical data map. The historical data map can intuitively display the relationship between historical data sequences of various attributes, making the interaction between different attributes clearer and providing strong data support for subsequent charging capacity prediction, thereby improving the utilization efficiency of charging facilities.

[0075] Furthermore, the implementation process of the data screening module also includes:

[0076] The historical data map is screened according to the first attribute coefficient and the second attribute coefficient; if the first attribute coefficient is greater than a predetermined first weight threshold, the sum of the first attribute coefficients of two corresponding nodes is calculated respectively, and the node with the largest sum of the first attribute coefficients is deleted; if the first attribute coefficient is less than or equal to the predetermined first weight threshold, the second attribute coefficient of the corresponding node is determined; if the second attribute coefficient is greater than the predetermined second weight threshold, the corresponding node is deleted; the historical data sequences corresponding to all retained nodes are spliced ​​to obtain time series key data.

[0077] In the embodiment of the present application, when predicting the future charging demand of charging station A in the highway service area, by screening the historical data map, redundant nodes that contribute less to the charging demand prediction can be removed, that is, important features are screened out, thereby improving the efficiency of charging quantity prediction; through threshold screening, it can be ensured that the remaining nodes have sufficient relevance and reliability, thereby reducing potential noise interference, so that the historical data can better reflect the changing trend of the charging amount; splicing the historical data sequences corresponding to all retained nodes helps to form a comprehensive and time-series key data set, which can better reflect the usage of the charging station, thereby improving the accuracy of predicting future charging demand.

[0078] Furthermore, the charging capacity prediction module constructs and trains a charging capacity prediction model based on the charging station spatial relationship diagram and the time series key data; inputs the time series key data within a preset time period before the prediction date into the charging capacity prediction model to obtain the future charging capacity of the target charging station on the prediction date; and determines whether to update the charging capacity prediction model in combination with the actual charging capacity of the target charging station, corresponding to Figure 1 Steps S4-S6 of Figure 4 The specific implementation process includes:

[0079] In the training phase, the time series key data of the historical time is divided to obtain training samples and labels; the training samples and the charging station spatial relationship diagram are input into the charging capacity prediction model to obtain the predicted charging capacity; the model parameters are optimized based on the labels and the predicted charging capacity to obtain the trained charging capacity prediction model;

[0080] In the prediction stage, the key time series data within the preset time period before the day to be predicted is obtained, and the key time series data is input into the trained charging capacity prediction model to obtain the future charging capacity of the day to be predicted;

[0081] In the model updating stage, the charging piles of the target charging station are scheduled according to the future charging amount and the charging pile distribution data; the actual charging amount of the target charging station is monitored in real time and compared with the future charging amount to determine whether to update the charging amount prediction model.

[0082] The embodiment of the present application can significantly improve the operational efficiency and resource allocation rationality of the charging station through the implementation of the above-mentioned charging capacity prediction module. The charging capacity prediction model constructed by combining the charging station spatial relationship diagram and time-series key data can accurately predict future charging needs; at the same time, the mechanism of updating the charging capacity prediction model ensures the continuous accuracy of the prediction results, allowing the charging station to flexibly respond to different usage situations and complex external environmental changes, further improving the overall charging service quality and user experience.

[0083] Furthermore, the charging capacity prediction model includes:

[0084] The spatial relationship diagram of the charging station and the key time series data are input into the feature extraction layer to obtain the spatial feature sequence and the time series feature sequence; through the spatiotemporal fusion layer, the spatiotemporal correlation of the spatial feature sequence and the time series feature sequence is captured, and the potential trends affecting the charging amount are extracted; combined with the fully connected layer and the prediction head, the future charging amount of the target charging station is obtained.

[0085] Specifically, the spatial characteristics reflect the spatial relationship between the target charging station and the surrounding charging stations and tourist destinations, while the temporal feature sequence reveals the temporal variation pattern of the charging amount. The spatiotemporal fusion layer can integrate the environmental information around the charging station (such as the number and model of charging piles at other charging stations and the passenger flow of surrounding tourist destinations) with the time series data (such as historical charging amount, user behavior characteristics and weather changes), so as to obtain the impact of the information around the charging station on the user's charging behavior.

[0086] According to the charging capacity prediction model, the future charging capacity demand and actual charging capacity data of charging station A in the highway service area are compared as an example. See Table 2 for a list of the future charging capacity and actual charging capacity of charging station A.

[0087] Table 2. Partial data on future charging capacity and actual charging capacity of charging station A in highway service area

[0088]

[0089] The spatial feature sequence and temporal feature sequence generated by the feature extraction layer in the embodiment of the present application can comprehensively reflect the operating environment of the charging station and the dynamic changes of the charging amount; the spatiotemporal fusion layer further captures the correlation between these features, so that the model can identify and extract the changing trend of the charging amount over time, which not only improves the accuracy of the charging amount prediction, but also enhances the model's adaptability to changes in complex scenarios.

[0090] Furthermore, the charging station spatial relationship diagram and the time sequence key data are input into the feature extraction layer to obtain the specific implementation process of the spatial feature sequence and the time sequence feature sequence:

[0091] Through the graph convolutional network, the spatial features of the target charging station in the charging station spatial relationship graph are extracted and converted into an adjacency matrix to obtain the spatial feature sequence; the time series key data is input into the convolutional neural network to capture the charging amount change characteristics, user charging behavior characteristics and weather change characteristics to obtain the time series feature sequence.

[0092] Specifically, the graph convolutional network uses the spatial relationship graph of charging stations to extract the spatial characteristics of charging station A in the highway service area and the surrounding charging stations, such as the mutual influence and potential demand between different stations, thereby reflecting the impact of the distance between adjacent charging stations and the number of charging piles on the demand for charging station A; and the convolutional neural network can capture the characteristics of charging amount changes, user charging behavior, weather changes, etc. in time-series key data, thereby obtaining the changing trend of charging demand over time.

[0093] The embodiment of the present application captures the spatial dependency between charging stations and the temporal changes in charging demand through the feature extraction layer, which can more accurately predict the charging demand and improve the prediction accuracy of the charging demand trend, provide data support for the reasonable configuration and optimization of charging station A, help avoid the problem of excess or shortage of resources, and improve service quality and resource utilization efficiency.

[0094] Furthermore, the space-time fusion layer includes:

[0095] The time series feature sequence is input into the bidirectional long short-term memory network to capture the dynamic change relationship of the historical data and obtain the first reconstruction feature. The specific calculation formula is expressed as:

[0096] ;

[0097] ;

[0098] in, represents the first reconstruction feature; They represent the hidden state data at the current time in the forward process and the backward process, respectively, which is the forward and backward time correlation of the time series feature sequence; represents the learning weight; represents the convolution operation; represents the long short-term memory network; Represents the input data at the current time, namely, the charging data, user behavior data and weather change data at the current time; Respectively represent the hidden state data of the previous time in the forward process and the backward process;

[0099] Through a dense convolutional neural network, the influence of different feature variables on the charging amount is extracted to obtain the second reconstruction feature; the spatial feature sequence and the second reconstruction feature are input into the cross-attention network to explore the dependency relationship between the charging stations and tourist attractions around the target charging station and the change in charging amount to obtain the spatiotemporal fusion feature.

[0100] Specifically, the bidirectional long short-term memory network can make full use of the forward and backward time correlation of historical data to capture the dynamic changes of charging demand over time; the dense convolutional neural network can deeply extract the different degrees of influence of various influencing factors (such as user behavior and weather changes) on the charging amount; the cross-attention network further combines spatial features and second reconstruction features, and by paying attention to the spatial relationship between charging stations, it fully explores the dependency relationship between the target charging station and surrounding charging stations and tourist destinations. It can not only obtain the impact of the spatial location of the charging station on the change of charging amount, but also capture the changing trend of charging demand caused by different dates (such as holidays, weekdays and weekends).

[0101] The embodiment of the present application can dynamically capture the changing trend of the historical data of the charging station through the spatiotemporal fusion layer, and accurately extract the influence of different characteristic variables on the charging amount, thereby improving the prediction accuracy of future charging demand. Through the cross-attention network, the model not only takes into account the changing trend of the time dimension, but also captures the spatial dependence of the distribution of charging stations, thereby improving the adaptability and prediction accuracy of the model in actual application scenarios.

[0102] Furthermore, the implementation process of the model updating phase includes:

[0103] The actual charging capacity of the target charging station is monitored in real time and compared with the future charging capacity to obtain a charging capacity error; if the charging capacity error is greater than a predetermined error threshold, the trained charging capacity prediction model is fine-tuned according to the actual charging capacity and the key timing data within the preset time. The specific formula is expressed as follows:

[0104] ;

[0105] ;

[0106] in, represents the model parameters after fine-tuning; represents the parameters of the original model; represents the learning rate; Express Find the gradient; A loss function representing the charge capacity prediction model; Indicates The predicted value of samples; No. The true value of samples; Represents the mean of all sample prediction values; Indicates the number of samples; represents the adjustment parameter, The value of Within the range and .

[0107] Specifically, for charging station A in the highway service area, the charging demand on the forecast day is obtained through the charging capacity prediction model. The daily charging capacity required can be purchased when the electricity price is low and stored in the energy storage battery to meet the charging demand on the forecast day; by comparing with the actual charging capacity, the charging capacity mutation date can be obtained, and the prediction model can be optimized in combination with the online update strategy, so that the prediction results are more in line with the actual situation. This real-time monitoring and feedback mechanism can effectively adjust the configuration of charging resources, thereby improving the operating efficiency of the charging station, optimizing the scheduling of the energy storage system, and reducing energy consumption and costs.

[0108] The embodiment of the present application can timely discover the deviation of the charge capacity prediction by real-time monitoring of the actual charge capacity of the target charging station and comparing it with the future charge capacity, thereby realizing the ability to dynamically adjust and optimize the prediction model. When the charge capacity error exceeds the predetermined threshold, the model parameters are fine-tuned according to the actual charge capacity and the key timing data within the preset time, which can not only optimize the operating efficiency of the charging station, but also effectively meet the charging needs of users, while improving the efficiency of resource allocation.

[0109] The embodiment of the present application realizes accurate prediction of charging demand for new energy vehicles in highway service areas by combining charging pile distribution, geographic information, historical charging data and meteorological data. The specific implementation process mainly includes the following steps: (1) Constructing a spatial relationship diagram of charging stations to accurately describe the spatial dependency relationship between the target charging station and surrounding charging stations and tourist destinations; (2) Through the charging capacity prediction model, deeply analyze the spatial relationship between charging stations and the time variation trend of charging demand; (3) Using real-time charging capacity monitoring and error feedback mechanism, timely adjust the parameters of the prediction model. The present invention proposes a charging station spatial relationship diagram, a charging capacity prediction model and a model update mechanism for the above three processes respectively; the construction of the spatial relationship diagram realizes the association modeling of the target charging station in the highway service area and the surrounding facilities, and improves the flexibility of charging demand prediction; the charging capacity prediction model further improves the ability to mine the influence of factors such as weather and holidays; the model update mechanism further improves the adaptability of the model and the stability of the prediction results.

[0110] Embodiment 2

[0111] In Example 1, the system of the present invention realizes the prediction of the future charging demand of the charging station A in the highway service area. In the embodiment of this application, the system proposed by the present invention will be described again to predict the future charging demand of the charging station to be built in the B highway service area. The specific implementation process is as follows:

[0112] The data acquisition module obtains the geographical distribution data of the charging stations to be built in the B highway service area, the charging data of historical time and the meteorological data; collects the vehicle images of historical time according to the high-definition cameras set at the entrance and exit of the highway service area; recognizes the vehicle images through image recognition technology to obtain the traffic flow data; the charging data of historical time is collected from the surrounding areas of the B highway service area The charging data of the charging stations in the service areas are averaged.

[0113] Furthermore, the data processing module cleans the charging data, meteorological data and vehicle flow data of the historical time to obtain historical charging data, historical meteorological data and historical vehicle flow data; and constructs a charging station spatial relationship diagram based on the geographical distribution data and the charging pile distribution data.

[0114] Furthermore, the specific implementation process of constructing the charging station spatial relationship diagram includes:

[0115] According to the geographical distribution data, determine the surrounding area of ​​the charging station to be built The coordinates of the adjacent charging stations and the adjacent tourist attractions in the area; obtaining the first mileage distance between the charging station to be built and the adjacent charging station, the second mileage distance between the adjacent tourist attraction and the charging station, and the charging pile data in the adjacent charging station; taking the charging station to be built, the adjacent charging station and the adjacent tourist attraction as nodes, combining the charging pile distribution data, the first mileage distance, the second mileage distance, the charging pile data in the station, the score of the adjacent tourist attraction and the proportion of tourists, calculate the edge weights between different nodes, and the specific formula is expressed as follows:

[0116] ;

[0117] ;

[0118] in, Indicates the charging station node to be built and the The edge weights between the adjacent charging station nodes, is a positive integer; Indicates The charging station node and The edge weights between the adjacent tourist destination nodes, the charging station to be built represents the Charging station nodes; Indicates the number of charging piles in the charging station to be built, which is determined by the surrounding The average number of charging piles in each service area charging station is obtained; Indicates The charging pile data of each of the adjacent charging stations; Indicates that the charging station to be built is a first mileage distance between the adjacent charging stations; The average of all first mileage distances; represents a natural constant; Indicates The ratings of the nearby tourist destinations and ; Indicates The proportion of tourists to the neighboring tourist destinations is normalized according to the average number of tourists per year of all the neighboring tourist destinations; Indicates The charging station node and The second mileage distance between the adjacent tourist destination nodes; represents the average value of the second mileage distance;

[0119] A spatial graph of the charging station to be constructed is established based on the node and edge weights; if the edge weight between the neighboring charging station node and the neighboring tourist destination node is less than or equal to the predetermined spatial weight, the corresponding edge in the spatial graph is deleted to obtain the spatial relationship graph of the charging station.

[0120] In order to further illustrate the role of the charging station spatial relationship diagram proposed in the present invention, the edge weights of the charging station node to be built in the B highway service area and the neighboring tourist destination node are compared as shown in Table 3.

[0121] According to the data in Table 3, it can be seen that the scores of neighboring tourist destinations, the proportion of tourists and the second mileage distance have a significant impact on the edge weight. For example, the score of LY02 tourist destination reaches full score. , the corresponding edge weight is also relatively large and reaches , which indicates that high-scoring tourist destinations attract more tourists, thereby increasing the edge weight between the charging stations to be built and the tourist destinations, reflecting that there may be more charging demand; on the contrary, the score of LY03 tourist destination is lower. The proportion of tourists is also relatively low. , and the second mileage is farther , so the edge weight is only This indicates that the lower the score of a tourist destination, the fewer tourists it can attract. The distance between the tourist destination and the charging station to be built is far, which means that most tourists will not choose to charge at the charging station to be built. Therefore, the second mileage distance, the score of the tourist destination, and the proportion of tourists can reflect the potential demand relationship between the charging station and the neighboring tourist destination.

[0122] Table 3. Edge weights of charging station nodes to be built in highway service area B and nodes of neighboring tourist destinations

[0123]

[0124] Furthermore, the data screening module obtains the similarity between historical data sequences of different attributes according to the historical charging data, the historical meteorological data and the historical vehicle flow data, and generates a historical data map; screens the historical data sequences by comparing the similarities, and determines the time series key data;

[0125] Furthermore, the charging capacity prediction module constructs and trains a charging capacity prediction model based on the charging station spatial relationship diagram and the time series key data; inputs the time series key data within a preset time period before the prediction date into the charging capacity prediction model to obtain the future charging capacity of the charging station to be built on the prediction date; and determines whether to update the charging capacity prediction model in combination with the actual charging capacity of the charging station to be built.

[0126] Specifically, for the charging stations to be built in the B highway service area, by constructing and training the charging capacity prediction model, it is possible to effectively integrate the spatial relationship diagram of the charging station and the key time series data, so as to accurately capture the various factors that affect the charging demand, such as geographical location, historical charging behavior and meteorological changes. Accurate charging capacity prediction can determine the peak and trough periods of charging demand in the B highway service area, so as to reasonably plan the number and type of charging piles to meet the changes in the flow of new energy vehicles. In addition, the charging capacity prediction results can also guide the efficient configuration of charging pile resources, such as opening more charging facilities or supercharging charging piles during periods of high demand to reduce user waiting time. At the same time, through dynamic monitoring and model update mechanisms, the configuration strategy of charging piles can be adjusted in time to ensure that the charging station can maintain a good service level in different time periods, thereby effectively improving the user experience and the overall utilization of charging facilities.

[0127] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A charging demand forecasting system for new energy vehicle charging stations based on big data, characterized in that: include: A data acquisition module acquires charging pile distribution data, geographical distribution data, historical charging data, meteorological data and vehicle images of target charging stations in high-speed service areas; and obtains vehicle flow data by combining image recognition technology with the vehicle images; The data processing module cleans the charging data, meteorological data and vehicle flow data to obtain historical charging data, historical meteorological data and historical vehicle flow data; obtains the spatial relationship between the target charging station and the adjacent charging stations and the adjacent tourist destinations according to the geographical distribution data; obtains the charging pile data in the adjacent charging station, the rating and the tourist ratio of the adjacent tourist destination, and constructs a charging station spatial relationship diagram in combination with the charging pile distribution data. The specific implementation process includes: According to the geographic distribution data, the coordinates of the neighboring charging stations and the neighboring tourist destinations in the area around the target charging station are determined; a first mileage distance between the target charging station and the neighboring charging station, a second mileage distance between the neighboring tourist destination and the charging station, and the charging pile data in the neighboring charging station are obtained, and a spatial graph of the target charging station is established; the spatial graph includes: spatial nodes and edges; the spatial nodes are the target charging station, the neighboring charging station, and the neighboring tourist destination; the edge weights between different spatial nodes are calculated by combining the charging pile distribution data, the first mileage distance, the second mileage distance, the charging pile data in the station, the rating of the neighboring tourist destination, and the proportion of tourists; if the edge weight between the neighboring charging station node and the neighboring tourist destination node is less than or equal to the predetermined spatial weight, the corresponding edge in the spatial graph is deleted to obtain the spatial relationship graph of the charging station; The data screening module obtains the historical data after data cleaning, calculates the first attribute coefficient and the second attribute coefficient between the historical data sequences with different attributes, and generates a historical data map; removes redundant attributes in the historical data through threshold screening to obtain time series key data; the implementation process includes: According to the historical charging data, the historical meteorological data and the historical vehicle flow data, a first historical data sequence set, a second historical data sequence set and a third historical data sequence set are respectively obtained, wherein the historical data sequence set contains the historical data sequences of different attributes; similarities between the historical data sequences of different attributes in the same historical data sequence set are obtained to obtain the first attribute coefficient; similarities between the historical data sequences in different historical data sequence sets are obtained to obtain the second attribute coefficient; the historical data map is generated by combining the historical data sequences of different attributes, the first attribute coefficient and the second attribute coefficient; the historical data map includes: nodes and edges; nodes are the historical data sequences; edges represent the relationship between different nodes, and edge weights are the first attribute coefficient and the second attribute coefficient; According to the first attribute coefficient and the second attribute coefficient, the historical data map is screened; if the first attribute coefficient is greater than a predetermined first weight threshold, the sum of the first attribute coefficients of two corresponding nodes is calculated respectively, and the node with the largest sum of the first attribute coefficients is deleted; if the first attribute coefficient is less than or equal to the predetermined first weight threshold, the second attribute coefficient of the corresponding node is determined; if the second attribute coefficient is greater than the predetermined second weight threshold, the corresponding node is deleted; the historical data sequences corresponding to all the retained nodes are spliced ​​to obtain time series key data; The charging capacity prediction module constructs and trains a charging capacity prediction model based on the charging station spatial relationship diagram and the time series key data of historical time; obtains the correlation between the charging capacity of neighboring charging stations, neighboring tourist destinations and date types through the model, and predicts the future charging capacity; monitors the actual charging capacity of the target charging station in real time, and compares it with the prediction result to determine whether to update the model; if necessary, fine-tunes the model parameters through the online update strategy; the specific implementation process includes: In the training phase, the time series key data of the historical time is divided to obtain training samples and true values; the training samples and the charging station spatial relationship diagram are input into the charging capacity prediction model to obtain the predicted charging capacity; the model parameters are optimized based on the true value and the predicted charging capacity to obtain the trained charging capacity prediction model; In the prediction stage, the key time series data within a preset time period before the day to be predicted is obtained, and the key time series data is input into the trained charging capacity prediction model to obtain the future charging capacity of the day to be predicted; In the model updating stage, the charging piles of the target charging station are scheduled according to the future charging amount and the charging pile distribution data; the actual charging amount of the target charging station is monitored in real time and compared with the future charging amount to determine whether to update the charging amount prediction model; The charging capacity prediction model specifically includes: The charging station spatial relationship graph and the time series key data are input into the feature extraction layer to obtain a spatial feature sequence and a time series feature sequence. The specific implementation process includes: extracting the spatial features of the target charging station in the charging station spatial relationship graph through a graph convolutional network, and converting them into an adjacency matrix to obtain the spatial feature sequence; inputting the time series key data into a convolutional neural network to capture the charging amount change characteristics, user charging behavior characteristics and weather change characteristics, and obtain the time series feature sequence; Through the spatiotemporal fusion layer, the spatiotemporal correlation of the spatial feature sequence and the temporal feature sequence is captured to extract the potential trend affecting the charging amount; the spatiotemporal fusion layer includes: The temporal feature sequence is input into a bidirectional long short-term memory network to capture the correlation between different date types and the dynamic change of charging amount, and obtain the first reconstruction feature; the influence of different attribute data features on the charging amount is extracted through a dense convolutional neural network to obtain the second reconstruction feature; the spatial feature sequence and the second reconstruction feature are input into a cross attention network to obtain the dependency between the charging stations and tourist destinations around the target charging station and the change of charging amount, and obtain the spatiotemporal fusion feature; Combining the fully connected layer and the prediction head, obtaining the future charging amount of the target charging station; The implementation process of the model updating phase includes: The actual charging capacity of the target charging station is monitored in real time and compared with the future charging capacity to obtain a charging capacity error; if the charging capacity error is greater than a predetermined error threshold, the trained charging capacity prediction model is fine-tuned according to the actual charging capacity and the key timing data within the preset time, combined with the online update strategy. The specific formula is expressed as follows: ； ； in, represents the model parameters after fine-tuning; represents the parameters of the original model; represents the learning rate; Express Find the gradient; A loss function representing the charge capacity prediction model; Indicates The predicted value of samples; No. The true value of samples; Represents the mean of all sample prediction values; Indicates the number of samples; represents the adjustment parameter, The value of Within the range and .

2. According to the big data-based new energy vehicle charging station charging demand prediction system of claim 1, it is characterized in that: Combining the image recognition technology with the vehicle image, the process of obtaining the vehicle flow data includes: According to the vehicle images, the vehicle images are screened by image recognition technology, and images of different vehicles are retained to obtain the total number of vehicles per day; in combination with the image segmentation algorithm, the license plate area of ​​the retained image is located and segmented to obtain the corresponding license plate of each vehicle image; based on the color analysis of the license plate image, the daily number of new energy vehicles is determined to obtain the proportion of new energy vehicles; wherein the traffic flow data includes the total number of vehicles and the proportion of new energy vehicles; the vehicle images are collected by high-definition cameras installed at the entrances and exits of the highway service area.

Images