Intelligent management system and method for mobile charging pile

Abstract

The invention relates to the technical field of charging pile management, in particular to an intelligent management system and method for a mobile charging pile, and the method comprises the steps: collecting data, fusing the operation data, user data and environment data of the mobile charging pile, and obtaining a normalized data set; inputting the normalized data set into a graph neural network for graph structure construction to obtain an initial graph structure, and fusing a feature matrix and an adjacent matrix in the initial graph structure to obtain a fused multi-dimensional data set; inputting the fused multi-dimensional data set into a demand prediction model to obtain a three-dimensional demand matrix; and determining a high-demand area, inputting the three-dimensional demand matrix and the area code of the high-demand area into a scheduling cost formula to obtain a scheduling cost, determining a charging pile calling priority, and generating a scheduling instruction according to the charging pile calling priority and the area code of the high-demand area. According to the scheme, the hybrid prediction model is constructed through the GCN and the LSTM, the prediction accuracy is improved, and the scheduling efficiency is optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of charging pile management, and particularly relates to a mobile charging pile intelligent management system and method. BACKGROUND

[0002] In actual operation, there are many problems in data collection of mobile charging piles. Positioning is easily affected by city high-rise buildings and electromagnetic interference, resulting in insufficient trajectory recording accuracy; current and voltage sensors have no redundancy design, so data collection is easily interrupted when a fault occurs, and instantaneous pulse interference affects the accuracy of the power curve. User data collection lacks a clear authorization mechanism, and sensitive information protection is insufficient, evaluation feedback processing efficiency is low, and user preferences are difficult to accurately extract. The frequency of environmental data acquisition is low, traffic and weather information is not updated in time, parking lot positioning error is large, and different regional data lacks fine statistics, which cannot provide effective support for demand analysis.

[0003] In terms of data processing, the fusion degree of multi-source heterogeneous data is low, the processing method of abnormal values and missing values is simple, which easily leads to data distortion; the regional demand difference and the actual state of the charging pile are not fully considered during scheduling, the resource matching efficiency is low, and the dynamic response to the power grid load is lacking, which easily affects the stability of power supply. SUMMARY

[0004] The present application improves the prediction accuracy and optimizes the scheduling efficiency by constructing a hybrid prediction model with GCN and LSTM.

[0005] The technical scheme provided by the present application is as follows: a mobile charging pile intelligent management method, the method comprising: Collecting mobile charging pile operation data, user data and environmental data, and fusing the mobile charging pile operation data, user data and environmental data through an edge computing node to obtain a normalized data set; Inputting the normalized data set into a graph neural network to construct a graph structure and obtaining an initial graph structure, and fusing the feature matrix and the adjacency matrix in the initial graph structure using a graph convolution network to obtain a fused multi-dimensional data set; Inputting the fused multi-dimensional data set into a demand prediction model to predict the charging demand probability distribution of different regions and different time periods, and obtaining a three-dimensional demand matrix; Determining a high demand area according to the three-dimensional demand matrix, inputting the three-dimensional demand matrix and the region code of the high demand area into a scheduling cost formula to obtain a scheduling cost, determining a charging pile calling priority according to the scheduling cost, and generating a scheduling instruction according to the charging pile calling priority and the region code of the high demand area.

[0006] Preferably, the specific process of collecting mobile charging pile operation data, user data and environmental data is as follows: The positioning module adjusts the collection frequency according to the moving speed of the mobile charging pile to collect and record the real-time time and timestamp, and collects real-time power through current and voltage sensors. The BMS dynamically adjusts the monitoring frequency according to the charging stage to monitor the remaining power of the mobile charging pile and sets a breakpoint resume mechanism; After the user authorizes, reservation information and evaluation feedback are collected, and differential privacy technology is used to process sensitive information; Connect to the traffic management department API to collect traffic volume, average speed, and congestion index of each road section, introduce road topology data, establish a road network distance matrix, set up a data caching mechanism, and collect parking space data through the geographic information system.

[0007] Preferably, the process of obtaining the normalized data set is as follows: Use 3 The outlier judgment threshold is set according to the criterion. If the value exceeds the threshold twice in a row, it is judged as an outlier and eliminated. The missing values ​​are filled by linear interpolation to obtain the cleaned data set. The min-max normalization formula is used to normalize the cleaned data, and the position data is converted into plane coordinates by Gaussian projection to obtain the normalized data set.

[0008] Preferably, the process of obtaining the initial graph structure is as follows: Define charging pile nodes, traffic section nodes, and parking lot nodes, obtain the node feature matrix, calculate the shortest path time of the road network using the Dijkstra algorithm, input the shortest path time into the edge weight calculation formula to calculate the edge weight, construct the edge set and adjacency matrix based on the edge weight, and construct the initial graph structure based on the node feature matrix, edge set, and adjacency matrix.

[0009] Preferably, the process of obtaining the fused multi-dimensional dataset is as follows: The adjacency matrix is ​​input into the first layer of the graph convolutional network to obtain the first-layer feature matrix. The attention weights are calculated by weighted summing the first-layer feature matrix, user preference data, and meteorological data through the attention mechanism to obtain a weighted feature matrix. The weighted feature matrix is ​​spliced ​​with the user preference data and meteorological data to obtain a fused feature matrix, and the splicing dimension is determined through 5-fold cross-validation. The fused feature matrix is ​​integrated with the original data timestamp to obtain a spatiotemporal structured dataset. The spatiotemporal structured dataset is standardized and packaged to obtain a fused multidimensional dataset.

[0010] Preferably, the process of obtaining the three-dimensional demand matrix is ​​as follows: The data time series processing is performed on the fused multi-dimensional data set, the fused feature matrix is sorted according to a geographical area and time, and a regional time series feature set is constructed; feature extraction is performed on the regional time series feature set and an adjacency matrix, the extracted features are input into a GNN spatial model, for the feature matrix of each time step, graph convolution calculation of GNN feature fusion consistency is adopted, a graph attention mechanism GAE is introduced, and a regional spatial feature sequence is obtained; the regional spatial feature sequence is input into an LSTM time series model, a residual connection is added between LSTM layers, and the demand probability of a single area and a single time step is obtained through full connection layer mapping to a charging demand probability distribution; and a three-dimensional demand matrix is obtained according to the demand probability of a single area and a single time step.

[0011] Preferably, the charging pile calling priority determination process is as follows: According to the three-dimensional demand matrix, the high power level demand probability sum of a certain area at a certain time step is calculated, if the high power level demand probability sum is greater than or equal to the regional demand threshold, the area is determined as a high demand area; the demand probability sum of the high demand area is input into a scheduling cost formula corresponding to the area code, the scheduling cost of each mobile charging pile is calculated, and the mobile charging piles are sorted according to the calling cost; the time required for each mobile charging pile to reach the high demand area is calculated, if the time required for the high-priority mobile charging pile is greater than 30 minutes, the next-priority mobile charging pile is selected.

[0012] Preferably, the scheduling instruction generation process is as follows: An elliptic curve encryption algorithm is used to generate charging pile public and private keys and public keys according to the charging pile equipment number, the area code and the power grid dispatching center public key, a registration transaction is submitted to a blockchain network, and identity authentication is completed after global network node consensus; an intelligent contract is written based on the Solidity language, in combination with the three-dimensional demand matrix regional demand threshold, the idle charging pile determination condition, the power grid load adjustment coefficient and the blockchain node list; the QueryIdle() function of the intelligent contract is called to filter out a charging pile list that is real-time task state idle and has an electric quantity greater than or equal to 70%, and the calling priority is determined through a scheduling cost formula; for a high demand area, a scheduling instruction is generated in combination with the area code and the charging pile expected arrival time.

[0013] The application further provides a mobile charging pile intelligent management system.

[0014] The application further provides a computer readable storage medium storing a computer program, and the computer program is executed by a processor to realize the mobile charging pile intelligent management method.

[0015] The application has the following beneficial effects: 1、Through the multi-dimensional collection of Beidou positioning, BMS system and environmental data, combined with 3σ criterion cleaning and min-max standardization processing, the overall improvement of data quality is realized. In the graph structure construction, the road network distance is introduced to replace the straight line distance to calculate the edge weight, and then the GCN and LSTM hybrid model is used to capture the regional spatial correlation and time regularity, so that the charging demand prediction error is reduced to within 8%. Accurate data support is provided for subsequent scheduling, avoiding resource mismatch or supply and demand imbalance.

[0016] 2、Based on the ECC encryption and PBFT consensus algorithm, the blockchain network is constructed to realize the credible evidence of charging pile identity authentication and scheduling instruction. The scheduling cost function (comprehensive distance and demand urgency) and the step power adjustment rule are integrated into the smart contract, which shortens the response time of high demand area by 30% and reduces the grid load fluctuation amplitude by 35%. At the same time, the introduction of private key multi-signature mechanism and third-party audit node effectively prevents data tampering and single point failure risk, and the scheduling full link delay is stable within 12 seconds, meeting the dual demands of real-time and security.

[0017] 3、From the dynamic frequency adjustment of data collection (such as reducing the positioning frequency when stationary to reduce redundancy), to the labeling and retention strategy of abnormal data, to the blockchain evidence and model iteration after scheduling execution, a complete closed-loop management is formed. On the user level, differential privacy technology protects sensitive information while improving the interactive experience through voice feedback and scoring system, and the satisfaction degree is improved from 72 points to 89 points; on the system level, the breakpoint resume mechanism reduces the data loss rate from 5.3% to 0.8%, and the resource utilization rate is improved by 37.8%, ensuring the stable operation of the charging network in complex scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 The flowchart of the mobile charging pile intelligent management method of the present application; Figure 2 The prediction process flowchart of the mobile charging pile intelligent management method of the present application. DETAILED DESCRIPTION

[0019] The following description is used to disclose the present application so that those skilled in the art can implement the present application. The preferred embodiments in the following description are only as examples, and other obvious modifications can be thought of by those skilled in the art. The basic principles of the present application defined in the following description can be applied to other embodiments, modifications, improvements, equivalents and other technical solutions without departing from the spirit and scope of the present application.

[0020] It is to be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the elements may be multiple, and the term "one" should not be understood as a limitation on the quantity.

[0021] like Figure 1 and Figure 2 As shown in the figure, the mobile charging pile operation data is collected through the built-in basic sensors, battery management system (BMS) and positioning module of the charging pile. Among them, the positioning module adopts the Beidou dual-mode positioning module and is equipped with an anti-interference antenna. The antenna adopts a multi-band design, which can effectively resist the interference of urban high-rise buildings and the signal interference of the electromagnetic environment, so that the positioning accuracy is improved to 0.5 meters. The collection frequency is dynamically adjusted according to the moving speed. When the speed is ≥30km / h, it is 0.5 seconds / time, and when the speed is ≥30km / h, it is 0.5 seconds / time. At 30 km / h, the system automatically switches to a 5-second monitoring interval. This not only records real-time location but also generates a detailed trajectory curve, facilitating subsequent analysis of the charging station's movement path and range. The current and voltage sensors feature a redundant design, with both primary and backup sensors operating simultaneously and a sampling frequency increased to 500 Hz. If the primary sensor fails (e.g., data fluctuates abnormally by more than ±10%), the system automatically switches to the backup sensor within 0.1 seconds, ensuring uninterrupted power data collection. A digital filtering algorithm is also implemented to automatically filter out transient pulse interference, resulting in a smooth real-time power curve. The BMS dynamically adjusts the monitoring frequency to monitor the remaining battery charge, starting with a 20-second monitoring interval during the initial charging phase (0-80% SOC) and a 60-second monitoring interval during the later stages (80%-100% SOC). The system monitors the temperature of various battery components (including the cell surface, battery pack casing, and heat sink) every 20 seconds, calculating the temperature difference between these components. If the temperature difference exceeds 5°C, a high-temperature warning is triggered and detailed temperature distribution data is uploaded. The system also monitors the voltage difference between individual cells. When it exceeds 0.1V, it is marked as abnormal and the data upload frequency is increased to 5 seconds per upload to ensure timely monitoring of battery balance. A breakpoint-resume transmission mechanism is implemented, caching data for up to 15 minutes during network interruptions and automatically retransmitting it upon network recovery. A mark-and-retain strategy is implemented for abnormal data, dynamically adjusting the 3σ threshold using the standard deviation of a sliding window (with a window size of 5 minutes) to avoid inadvertent deletion of normal fluctuations during startup.

[0022] The data is directly stored in the local database, all running data is encrypted by AES-256 encryption algorithm before being stored in the encrypted database of the local edge node, the encryption key is updated automatically every 24 hours, and the key is stored and managed by the hardware security module (HSM) to prevent key leakage. In addition to the conventional check, the data integrity check introduces the hash value comparison, which calculates the hash value of the stored data every hour and compares it with the original hash value to ensure that the data has not been tampered with. After the charging pile is started, each sensor and module collects data at a set frequency, which is processed and encrypted and then stored in the local encrypted database. The system monitors the equipment running state in real time, and when an exception occurs (such as sensor failure, data anomaly), the corresponding early warning mechanism and data supplement strategy are automatically triggered. Data integrity check is performed every hour to ensure the accuracy and reliability of the data. Thus, the anti-interference position trajectory data with millisecond-level timestamp, the redundant collected filtered real-time power curve, and the battery state data set with temperature gradient monitoring and warning identification are obtained.

[0023] When collecting user data, the data collection authorization agreement needs to be clearly displayed when the user first uses the APP. The agreement specifies the type of data collected, its purpose, and the storage period. Only after the user's authorization can non-essential preference data be collected. When the user initiates a reservation, in addition to recording the reservation time, expected charging amount, and designated charging location, the user's historical charging data for the past three months is also retrieved simultaneously. Through big data analysis, the fast / slow charging preference (such as 70% or more choosing fast charging, marked as "fast charging preference user"), average charging time, and common payment method are obtained and stored in association with the reservation information. The APP has a voice feedback function, users can submit evaluations through voice, and the system uses deep learning-based speech recognition technology to accurately convert speech into text. The text evaluation is used together with natural language processing technology to extract keywords (such as "slow charging" "loose interface" "good service"), and is stored in categories such as "facility problem" "service quality" "charging experience". At the same time, a rating entry of 1-5 stars is set on the charging completion page to facilitate users to quickly feedback the charging experience.

[0024] When analyzing user historical charging data, differential privacy technology is used to process sensitive information, such as blurring the average charging fee by ±10% and dividing the specific charging time into time periods (e.g., "8:00-10:00") to avoid revealing user's accurate consumption information and travel patterns. Regularly (once a month), a data usage report is pushed to the user, detailing the data usage scenarios and scope, and the user can modify the data collection range or delete part of the historical data at any time. After the user registers for the APP and authorizes data collection, the system collects reservation information, order data, and feedback information in real time during user use. After the collected data is processed and classified, it is stored in the user preference database. A data usage report is generated and pushed to the user every month, and the data collection strategy is adjusted based on user feedback. Thus, the user's order data containing reservation information, user portrait with differential privacy processing and preference tags, and multi-channel classified user feedback data set are obtained with the user's authorization.

[0025] When collecting environmental data, the real-time traffic data API of the traffic management department is connected, and the data acquisition frequency is increased to every 3 minutes. The information obtained includes vehicle flow, average speed, congestion index (0-10 level) of each road section, etc. The traffic conditions within 1 km around the charging pile are highlighted, and when the congestion index exceeds 7, it is automatically marked as a congestion area. Road topology data is introduced to establish a road network distance matrix instead of straight-line distance. Real-time road condition data is obtained through OpenStreetMap, and the actual travel time between nodes is calculated as the basis for edge weight calculation. At the same time, a data caching mechanism is set up, when the network is interrupted, automatically enable cache data (cache time not more than 15 minutes), and after the network is restored, quickly synchronize the latest data to ensure the continuity of the data.

[0026] Access the real-time data interface of the meteorological service platform, in addition to collecting temperature, humidity, precipitation, wind force and other parameters, add air quality index (AQI) and visibility parameters. When detecting heavy rain (daily precipitation ≥50mm), strong wind (wind force ≥6), AQI ≥200, etc. When the weather or environmental conditions are bad, automatically increase the data refresh frequency to 1 minute / time, and trigger the bad weather warning mark.

[0027] Parking lot data is associated with geographic information system (GIS) basic data, and the location of empty parking spaces is accurately determined to specific coordinates (error ≤3 meters). Combined with the channel layout of the parking lot and the size of the charging pile, the optimal parking path is analyzed to facilitate the navigation of the mobile charging pile to the empty parking space. At the same time, the total number of parking spaces and the number of occupied parking spaces are synchronized every 5 minutes, the parking occupancy rate is calculated, and a heat map is generated to visually display the parking space tightness of each area of the parking lot.

[0028] During the spatial matching of environmental data, buffer zone analysis is introduced. Buffer zone analysis is performed, centering the current charging station location, to collect data for 500-meter, 1-kilometer, and 2-kilometer radiuses. Data such as traffic flow distribution, meteorological parameter differences, and parking occupancy rate changes within each radius are analyzed. This provides refined data support for charging demand analysis within different radiuses. The system regularly connects to various data interfaces to acquire traffic, meteorological, and parking lot data. After processing, correlation, and analysis, it is categorized and stored by region and buffer zone. When special conditions such as severe weather or traffic congestion are detected, appropriate data processing and early warning mechanisms are activated to ensure the timeliness and validity of the data. This results in a traffic condition dataset with a caching mechanism and regional identifiers, meteorological data with severe weather warnings including air quality and visibility, and parking lot occupancy data with heat maps and associated geographic coordinates.

[0029] Deploy edge computing nodes at charging pile terminals to fuse the collected multi-source heterogeneous data. The collected data includes location trajectory data ,in , For the The longitude of the record (unit: degrees), For the The latitude of the record, is the timestamp (unit: millisecond); power curve data ,in For the Recorded charging power (unit: kW); battery status data ,in , For the The remaining battery power of the record (unit: %), is the timestamp (unit: milliseconds), is the battery temperature (unit: °C); user order data ,in , is the reservation time (unit: seconds), is the charging time (unit: seconds), is the charging amount (unit: kWh); traffic data ,in , For the Congestion index of the record (range 0-10), is the average vehicle speed (unit: km / h); meteorological data ,in , is the temperature (unit: °C), is the humidity (unit: %), Air quality index (range 0-500); parking lot data ,in , is the parking space occupancy rate (unit: %), is the parking space coordinate (unit: degree).

[0030] Use 3 Criteria, define the outlier judgment threshold For the data series , calculate the mean ( is the sequence length) and the standard deviation ,when If the value appears more than 2 times in a row, it will be considered as an outlier and removed to avoid misjudgment of a single fluctuation. Linear interpolation is used to fill in missing values: and ( ) ( ),but After cleaning, we get the dataset , where the superscript “ " represents the cleaned data. The min-max normalization formula is used to normalize the cleaned data. The formula is ,in, is the minimum value of the data sequence, is the maximum value of the data sequence, after normalization . For location data Use Gauss projection to transform into plane coordinates (Unit: meter), obtain normalized data set .

[0031] Build the graph structure based on the normalized data set. Define the node set ,in, For charging pile nodes (number ), the eigenvector , is the normalized charge, is the normalized power, is the normalized temperature; is the traffic section node (number ), the eigenvector , is the normalized congestion index, is the normalized vehicle speed; For parking lot nodes (number ), the eigenvector , is the normalized occupancy rate, the node feature matrix where is the feature dimension feature vector is zero-padded to 3 dimensions. The node feature matrix is obtained. The edge weight is calculated using the road network travel time, and the edge weight calculation formula is where, is the time decay coefficient, is the actual travel time from node to (the shortest path time of the road network is calculated by Dijkstra algorithm), when , , otherwise , so as to obtain the adjacency matrix where . According to the node feature matrix and the adjacency matrix , the graph structure is constructed, where the edge set , and the initial graph structure is obtained.

[0032] The feature matrix and the adjacency matrix in the initial graph structure are fused by using the graph convolution network (GCN). The GCN hyperparameters are determined by grid search, the first layer hidden unit number is set to 128, the second layer is set to 64, and the dropout rate is 0.3 to avoid overfitting. The first layer of the graph convolution network (GCN) is calculated as where, is the identity matrix (adding a self-loop), is the degree matrix of , satisfying , is the weight matrix, is the bias term, is the ReLU activation function, that is, , and the first layer feature matrix is obtained. The attention weight is calculated by weighted summation of the first layer feature matrix , user preference data and weather data through the attention mechanism, where, is the attention parameter, represents vector splicing, is the activation function, that is, . The weight is adjusted in combination with user preference and weather factors, where , , is the influence coefficient. Obtain the weighted feature matrix , .

[0033] The GCN features (weighted feature matrix ) are spliced with user preference data , weather data , and the calculation formula is , is a 32-dimensional vector, is a 32-dimensional vector. Obtain the fusion feature matrix . The splicing dimension is determined by 5-fold cross-validation, and the prediction error of 128-dimensional features on the validation set is reduced by 11.7% compared to 64-dimensional features. The fusion feature matrix is integrated with the original data timestamp (unit: milliseconds), grouped by geographical area and time window (i.e. 300 seconds), and the feature set of the area at time is defined. Obtain the spatio-temporal structured dataset . The spatio-temporal structured dataset is standardized and packaged, and each record format is , is the unique identifier of the node, is the geographical area code, is the start time of the time window, is the original data association index. Obtain the fused multi-dimensional dataset .

[0034] The fused multi-dimensional dataset is processed by a demand prediction model to obtain a “time-area-power” three-dimensional demand matrix, and the demand prediction model is constructed by mixing graph neural network (GNN) and long short-term memory network (LSTM). The fused multi-dimensional dataset is processed in time sequence, the fusion feature matrix is sorted by geographical area and time , and the area time sequence feature sequence is constructed, wherein is the number of time windows of the area , satisfying , is consistent with the time window generated by the multi-dimensional dataset. Obtain the area time sequence feature set . The area time sequence feature set and the adjacency matrix Feature extraction is performed. Both are input into the GNN spatial model, and the feature matrix of each time step is calculated using GNN feature fusion consistent graph convolution wherein is a weight matrix, is a bias term, is a ReLU activation function. A graph attention mechanism GAE is introduced, and dynamic weights are given to adjacent region features in graph convolution. The expression ability of spatial correlation is enhanced through a multi-head attention mechanism, and the number of heads is set to 4, and the output dimension of each head is 16. A region spatial feature sequence is obtained, wherein .

[0035] The region spatial feature sequence is input into the LSTM time series model, and the of each region is input into a 3-layer LSTM network (referring to the standard LSTM structure of Keras), the number of hidden units in the first layer is , a sequence is returned, the dropout rate is , the number of hidden units in the second layer is , a sequence is returned, the dropout rate is , the number of hidden units in the third layer is , no sequence is returned, and a timing feature is output . A residual connection is added between the LSTM layers to alleviate the gradient disappearance problem of the deep network, and the residual connection ratio is set to 0.5. A region timing feature set is obtained. Demand probability distribution prediction is performed according to the region timing feature set . The charging demand probability distribution is mapped through a fully connected layer wherein ( = 10, corresponding to the power level interval of 0-10 kWh), is a bias term, is a temperature coefficient, when , the difference of high value features is amplified, and the peak value of the high probability interval is more prominent, the function ensures that the probability sum is 1. A single-region single-time-step demand probability is obtained. The steps of feature extraction to demand probability distribution prediction are repeated using a sliding time window (step ) to obtain all region demand probabilities for time steps (one day) in the future. A multi-region multi-time-step demand probability set is obtained. Finally, a "time-region-power" three-dimensional demand matrix is obtained, wherein represents the demand probability of the time steps ( ), No. Region, The demand probability of each power level.

[0036] By using GNN to capture spatial correlations between regions (such as the linkage between charging demand in commercial areas and surrounding residential areas), and combining it with LSTM to explore temporal patterns (the periodicity of morning and evening peaks), the prediction error is reduced to less than 8%. Ablation experiments show that the MAE of the GNN model alone is 12.3%, the LSTM model alone is 10.7%, and the hybrid model is reduced to 7.9%, verifying the effectiveness of spatiotemporal feature fusion. For example, the actual demand for 6-8kWh in area 10 at 18:05 accounted for 22%, while the predicted value was 20%, with an error of only 2%. Input multiplexing and fusion of multi-dimensional datasets ,avoiding data conversion loss, such as the spatial features of area 10 directly associated with the traffic congestion data fused by edge computing, making the prediction more accurate, the three-dimensional demand matrix This provides a basis for scheduling. For example, based on the high demand forecast for area 10, three idle charging piles are dispatched from area 9 (low demand) in advance, increasing the response speed by 30%.

[0037] For example, a city is divided into =50 areas (encoding =1 to =50), each region generates a fusion feature data every 5 minutes. =10 (commercial area) as an example, Included in =1620000000000 (May 1, 2021 08:00:00) feature vector =[0.62,0.35,...,0.81] (128 dimensions, including normalized features such as charging pile power and surrounding traffic flow). =10 (commercial area) will be open from 08:00 to 18:00 on May 1, 2021 = 120 time windows (one every 5 minutes), ,in Corresponding to 08:05. (08:00), Area of With adjacent areas (residential area), The characteristics of (office area) are obtained through the adjacency matrix Association (assuming (Detour distance 2.3km), (Direct distance 1.8km)). After graph convolution calculation, (64 dimensions, filter invalid features, strengthen the spatial correlation feature of "commercial area 08:00 charging pile demand increase"). Area =10 The spatial features containing 120 time steps, the LSTM network learns the time regularity of "workday 08:00-10:00, 18:00-20:00 charging demand peak", and outputs =[0.78, 0.12,..., 0.65] (16 dimensions, highlight the morning and evening peak timing features). Area =10 After calculation by the full connection layer, we get =1620042000000 (18:05) demand probability =[0.02, 0.03, 0.05, 0.10, 0.15, 0.20, 0.18, 0.12, 0.08, 0.07], indicating that the probability of 6-8kWh power demand is the highest (20%), which is consistent with the regularity of concentrated power supply for vehicles in the evening peak in the commercial area. For area =10 sliding calculation, we get the demand probability sequence from 18:05 to 18:00 the next day, where the high power demand probability from 18:05 to 20:00 is relatively high, while the demand probability from 02:00 to 06:00 in the morning is concentrated in the low power level (0-2kWh, probability total 35%) indicates that the demand probability of the 1st time step (18:05), area 10, and 6-7kWh power level is 20%.

[0038] According to the charging pile device number , area code (the same as the area code of the three-dimensional demand matrix), and the power grid dispatching center public key , the charging pile private key and public key are generated by the elliptic curve encryption algorithm (ECC), and the registration transaction containing , , is submitted to the blockchain network, and the identity authentication is completed after the network node consensus (using the practical Byzantine fault tolerance algorithm PBFT, consensus delay ≤500ms). The blockchain energy coordination network is constructed, and the nodes include 50 charging pile clusters of the area (20 in each area, a total of 1000), 1 power grid dispatching center node, and 3 supervision nodes. Obtain the blockchain node list and network topology structure.

[0039] According to the area demand threshold of the three-dimensional demand matrix (the sum of the area demand probability High demand when idle charging pile is used (no charging task for 30 consecutive minutes and the power is 70%), grid load adjustment coefficient ( ), blockchain node list , write smart contracts based on Solidity language, including demand judgment rules (if , the area is a high-demand area, triggering the charging pile scheduling instruction), scheduling priority (high-demand areas get idle resources first, adjacent areas The scheduling weight of idle charging piles , is the distance from the center of the area, in km), power adjustment rules (charging power ,in (Maximum output power of charging pile), (Load impact coefficient). The contract is deployed after verification by the blockchain node, and the address is ContractAddr. The smart contract calls the QueryIdle() function to retrieve idle charging piles, and based on the idle judgment conditions, it filters out the real-time task status of the charging piles. And the real-time power of the charging pile A list of devices, sorted by the scheduling priority in the contract Sort.

[0040] The priority of charging piles is determined by the scheduling cost formula, and the sorted scheduling priorities are reordered. The scheduling cost formula is: , For charging piles Current location to target area The road network distance, It is a special penalty item for charging piles. Based on historical scheduling records, if If the success rate of the last three scheduling executions is less than 90%, then ,the remaining , is the penalty term weight coefficient, The higher the cost, the lower the priority. , generate scheduling instructions ,in The coding of high-demand areas is consistent with the coding of the three-dimensional demand matrix area. ( It is a dynamic value. , when smooth ), The secondary scheduling plan is triggered when the suboptimal but faster-responding charging pile is selected. The timestamp of the time when the scheduling instruction is triggered. The timestamp when the charging pile is expected to arrive at the target area.

[0041] According to the real-time load of the power grid 、Current power of charging pile , power adjustment rules, calculate the load adjustment coefficient based on contract rules When the load suddenly increases by more than 10%, a step-by-step power reduction strategy is adopted, reducing the power by 20% in the first minute and then reducing it by 10% in the second minute to avoid power grid shock. When the power is reduced to ,when When, allow . Get dynamic charging power parameters According to the blockchain node list , dynamic charging power parameters , scheduling instructions , through the charging pile private key After signing, it is sent to the corresponding node and the execution result is (Success / failure) upload to the blockchain for storage, smart contract based on Update Status (Success) or trigger secondary scheduling. Obtain the instruction execution record stored in the regional chain. Record the entire scheduling link delay. If the timestamp difference between the demand trigger and the charging pile response exceeds 15 seconds, it will automatically be marked as an abnormality, triggering the node performance optimization process.

[0042] For example, at a certain moment 1620042000000 (18:05, corresponding to time steps), region The fragment of the three-dimensional demand matrix (commercial area) is , real-time load of the power grid , .area Charging piles (No. 101) generated and , register to the blockchain and obtain the node ID , and the same region and grid nodes The subnet is formed to provide a node basis for the subsequent execution of the scheduling instructions in the area. of , hour , meeting high demand conditions. Based on the contract Calculation rules, area of ( ) takes precedence over region of ( generating instructions .

[0043] The processes described above with reference to the flowcharts can be implemented as computer software programs in accordance with embodiments of the present disclosure. Embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program comprising program code for executing the methods illustrated by the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network by a communication section, and / or installed from a detachable medium. When the computer program is executed by a central processing unit (CPU), the above-described functions defined in the methods of the present application are executed. It should be noted that the computer readable medium described above in the present application can be a computer readable signal medium or a computer readable storage medium or any combination of the two. The computer readable storage medium may, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination of the above. More specific examples of the computer readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus or device. In the present application, the computer readable signal medium can include a data signal carried in a baseband or as a part of a carrier wave, in which a computer readable program code is carried. Such a propagated data signal can take on many forms, including but not limited to, an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer readable signal medium can also be any computer readable medium that can send, propagate or transfer a program for use by or in connection with an instruction execution system, apparatus or device. The program code contained on the computer readable medium can be transmitted by any suitable medium, including but not limited to, wireless, wire, optical cable, RF, or any suitable combination of the above.

[0044] The computer program product of the present application can be a computer program product comprising a computer-readable medium bearing computer program code embodied therein for use with a computer. The computer program code can be code defining and / or implementing the present application. The computer program code can be organized into one or more computer program components, including but not limited to, program components for implementing the features of the present application. The one or more computer program components can each include the code implementing the features of the present application. The one or more computer program components can each be embodied on the computer-readable medium. The computer-readable medium can be a tangible computer-readable medium. In operation, the computer program code is provided on the computer-readable medium to a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the code can be executed to implement the present application.

[0045] It will be appreciated by persons skilled in the art that the present application is not limited to what has been described above and illustrated in the drawings. The application is defined by the claims and many modifications and variations of the illustrative embodiments can be made by those skilled in the art without departing from the scope of the present application, which is defined by the following claims.

Claims

1. A mobile charging pile intelligent management method, characterized in that: The method comprises: Collect mobile charging pile operation data, user data, and environmental data, and fuse them through edge computing nodes to obtain a normalized data set; The normalized dataset is input into the graph neural network to construct the graph structure and obtain the initial graph structure. The feature matrix and adjacency matrix in the initial graph structure are fused using the graph convolutional network to obtain the fused multi-dimensional dataset. The fused multi-dimensional data set is input into the demand forecasting model to predict the probability distribution of charging demand in different regions and time periods, and obtain a three-dimensional demand matrix; Determine high-demand areas based on the three-dimensional demand matrix, input the three-dimensional demand matrix and the area codes of the high-demand areas into the scheduling cost formula to obtain the scheduling cost, determine the charging pile call priority based on the scheduling cost, and generate scheduling instructions based on the charging pile call priority and the area codes of the high-demand areas.

2. A mobile charging pile intelligent management method according to claim 1, characterized in that: The specific process of collecting mobile charging pile operation data, user data and environmental data is as follows: The positioning module adjusts the collection frequency according to the moving speed of the mobile charging pile to collect and record the real-time time and timestamp, and collects real-time power through current and voltage sensors. The BMS dynamically adjusts the monitoring frequency according to the charging stage to monitor the remaining power of the mobile charging pile and sets a breakpoint resume mechanism; After the user authorizes, reservation information and evaluation feedback are collected, and differential privacy technology is used to process sensitive information; Connect to the traffic management department API to collect traffic volume, average speed, and congestion index of each road section, introduce road topology data, establish a road network distance matrix, set up a data caching mechanism, and collect parking space data through the geographic information system.

3. A mobile charging pile intelligent management method according to claim 2, characterized in that: The process of obtaining the normalized data set is as follows: Use 3 The outlier judgment threshold is set according to the criterion. If the value exceeds the threshold twice in a row, it is judged as an outlier and eliminated. The missing values ​​are filled by linear interpolation to obtain the cleaned data set. The min-max normalization formula is used to normalize the cleaned data, and the position data is converted into plane coordinates by Gaussian projection to obtain the normalized data set.

4. A mobile charging pile intelligent management method according to claim 3, characterized in that: The process of obtaining the initial graph structure is as follows: Define charging pile nodes, traffic section nodes, and parking lot nodes, obtain the node feature matrix, calculate the shortest path time of the road network using the Dijkstra algorithm, input the shortest path time into the edge weight calculation formula to calculate the edge weight, construct the edge set and adjacency matrix based on the edge weight, and construct the initial graph structure based on the node feature matrix, edge set, and adjacency matrix.

5. A mobile charging pile intelligent management method according to claim 4, characterized in that: The process of obtaining the fused multi-dimensional dataset is as follows: The adjacency matrix is ​​input into the first layer of the graph convolutional network to obtain the first-layer feature matrix. The attention weight is calculated by weighted summing the first-layer feature matrix, user preference data, and meteorological data through the attention mechanism to obtain a weighted feature matrix. The weighted feature matrix is ​​concatenated with the user preference data and meteorological data to obtain a fused feature matrix. The concatenation dimension is determined through 5-fold cross-validation. The fused feature matrix is ​​integrated with the original data timestamp to obtain a spatiotemporal structured dataset; the spatiotemporal structured dataset is standardized and packaged to obtain a fused multidimensional dataset.

6. A mobile charging pile intelligent management method according to claim 5, characterized in that: The process of obtaining the three-dimensional demand matrix is ​​as follows: The fused multi-dimensional data set is processed in a temporal manner, and the fused feature matrix is ​​sorted by geographical region and time to construct a regional temporal feature set. Feature extraction is performed on the regional temporal feature set and the adjacency matrix, and the extracted features are input into the GNN spatial model. For the feature matrix of each time step, a graph convolution calculation consistent with the GNN feature fusion is used, and the graph attention mechanism GAE is introduced to obtain the regional spatial feature sequence. The regional spatial feature sequence is input into the LSTM time series model, and residual connections are added between LSTM layers. The charging demand probability distribution is mapped to the fully connected layer to obtain the demand probability of a single region and a single time step. Based on the demand probability of a single region and a single time step, a three-dimensional demand matrix is ​​obtained.

7. A mobile charging pile intelligent management method according to claim 6, characterized in that: The charging pile call priority determination process is as follows: Based on the three-dimensional demand matrix, the total probability of high-power level demand in a certain area at a certain time step is calculated. If the total probability of high-power level demand is greater than or equal to the regional demand threshold, it is determined to be a high-demand area. The corresponding area code of the total demand probability of the high-demand area is input into the scheduling cost formula to calculate the scheduling cost of each mobile charging pile. All mobile charging piles are prioritized according to the call cost. Calculate the time required for each mobile charging station to reach the high-demand area. If the time required for a high-priority mobile charging station is greater than 30 minutes, select the next-priority mobile charging station.

8. A mobile charging pile intelligent management method according to claim 7, characterized in that: The scheduling instruction generation process is as follows: Using the elliptic curve cryptography algorithm, the charging pile's public and private keys and public keys are generated based on the charging pile device number, regional code, and the public key of the power grid dispatch center. A registration transaction is submitted to the blockchain network, and identity authentication is completed after consensus among all nodes in the network. Based on the Solidity language, a smart contract is written, combining the regional demand threshold of the three-dimensional demand matrix, the idle charging pile judgment criteria, the power grid load adjustment coefficient, and the blockchain node list. The smart contract calls the QueryIdle() function, which filters out a list of charging piles with a real-time task idle status and a power level ≥ 70% based on the idle judgment criteria, and determines the call priority using the scheduling cost formula. For high-demand areas, dispatch instructions are generated by combining the area code and the estimated arrival time of the charging station.

9. A mobile charging pile intelligent management system, characterized in that: The system is used to execute the mobile charging pile intelligent management method described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the mobile charging pile intelligent management method according to any one of claims 1 to 8.

Images