Multi-scene adaptation method and system for charging network

Abstract

The invention relates to the technical field of electric data processing, and provides a multi-scene adaptation method and system for a charging network, and the method comprises the steps: obtaining an environment parameter sequence and a behavior parameter sequence of a plurality of charging nodes in a target region, carrying out the feature extraction to construct a complex adaptation map, carrying out the multi-dimensional feature embedding, and obtaining a map embedding matrix, and performing stability analysis on the atlas embedding matrix to obtain a multi-level adaptation atlas, determining an adaptation scene cluster to which the current to-be-adapted node belongs based on the multi-level adaptation atlas, and dynamically adjusting a charging scheduling strategy and a parameter configuration scheme of the current to-be-adapted node according to a center feature vector of the adaptation scene cluster. Through adaptive scene cluster identification and charging strategy adjustment based on the node feature vector, the efficiency and resource utilization of the charging network are improved, and the problems that the prediction and scheduling efficiency of the charging demand is low, the flexibility is poor, and the charging resources are not fully utilized in the face of complex adaptive demands are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric data processing, and in particular to a multi-scene adaptation method and system for a charging network. BACKGROUND

[0002] With the popularization of electric vehicles (EVs), charging infrastructure has become a core component of supporting green travel. As an important part of the electric vehicle industry, the charging network not only involves the deployment of charging piles, but also involves efficient matching and charging scheduling between electric vehicles and charging piles. How to provide flexible, efficient and intelligent charging services according to different application scenarios (such as cities, suburbs, highways, etc.) has become a key problem in the current construction of electric vehicle charging networks.

[0003] In related technical means, the scene adaptation of the charging network mainly relies on static data mapping and rule matching, and the potential demand of the charging node is identified based on geographic information system (GIS) regional analysis, and then the layout of the charging pile and the allocation of the charging power are determined. In addition, there is also an analysis of the charging behavior mode of the user to optimize the load balancing of the charging station, a pattern recognition algorithm based on historical data and fixed scheduling rules, and the use of these data to predict demand and make charging plans according to demand conditions, which can realize the matching of charging demand and the reasonable allocation of load.

[0004] For the above technical solution, although the static pattern matching and demand prediction algorithm based on historical data can realize the basic management and scheduling of the charging station, when facing complex adaptation requirements in different geographical locations, climate changes, user behavior differences and other multi-scene conditions, it is difficult to fully consider the dynamic changes of the environment of the charging node, and it is not possible to deeply analyze the fusion of charging behavior and environmental characteristics, resulting in low efficiency and poor flexibility in predicting and scheduling charging demand during peak periods or special environments, leading to overloading or idling of some charging nodes, and not fully utilizing charging resources. SUMMARY

[0005] In order to improve the problem of low efficiency and poor flexibility in predicting and scheduling charging demand when facing complex adaptation requirements, and not fully utilizing charging resources, the present application provides a multi-scene adaptation method and system for a charging network.

[0006] The application provides a multi-scene adaptation method of a charging network, comprising: acquiring environmental parameter sequences and behavior parameter sequences of multiple charging nodes in a target area, performing feature extraction on the environmental parameter sequences and the behavior parameter sequences to obtain environmental response vectors and behavior response vectors; constructing a composite adaptation graph by using the environmental response vectors and the behavior response vectors, performing multi-dimensional feature embedding on the composite adaptation graph to obtain a graph embedding matrix, performing stability analysis on the graph embedding matrix to obtain a node scene adaptation factor set; performing sequence clustering on the node scene adaptation factor set to generate a multi-level adaptation graph, determining an adaptation scene cluster to which a current to-be-adapted node belongs based on the multi-level adaptation graph, wherein the current to-be-adapted node refers to a node with a specific geographic location, user demand characteristics and historical charging data; and dynamically adjusting a charging scheduling strategy and a parameter configuration scheme of the current to-be-adapted node according to a center feature vector of the adaptation scene cluster.

[0007] As a preferred solution, the step of acquiring environmental parameter sequences and behavior parameter sequences of multiple charging nodes in a target area, and performing feature extraction on the environmental parameter sequences and the behavior parameter sequences to obtain environmental response vectors and behavior response vectors, comprises: collecting environmental parameter sequences and behavior parameter sequences of multiple charging nodes in a target area through a preset sensor and a communication module; performing periodic segmentation processing and local fluctuation calculation on the environmental parameter sequences to obtain a node terrain variation index, a climate cycle mode and a temperature difference distribution coefficient, performing window weighting processing and abnormal point elimination on the behavior parameter sequences to obtain a user stay time dynamic weight, a charging request density matrix and a rhythm variation curve; combining and encoding the node terrain variation index and the climate cycle mode to obtain a first environmental representation vector, normalizing and fusing the temperature difference distribution coefficient and the user stay time dynamic weight to obtain an environmental behavior cross factor, and convoluting and fitting the charging request density matrix and the rhythm variation curve to obtain a user behavior spectrum vector; performing feature splicing on the first environmental representation vector and the environmental behavior cross factor to obtain an environmental response vector, and performing feature residual synthesis on the environmental behavior cross factor and the user behavior spectrum vector to obtain a behavior response vector.

[0008] As a preferred solution, the step of constructing a composite adaptation graph by using the environment response vector and the behavior response vector, performing multi-dimensional feature embedding on the composite adaptation graph to obtain a graph embedding matrix, and performing stability analysis on the graph embedding matrix to obtain a node scene adaptation factor set comprises: constructing a node environment graph based on the environment response vector, constructing a node behavior graph based on the behavior response vector, performing graph-level joint fusion on the node environment graph and the node behavior graph to generate a composite adaptation graph; performing nested multi-view feature extraction on the composite adaptation graph to obtain a node relationship tensor and an edge weight weight matrix, performing tensor compression decomposition on the node relationship tensor to obtain a node similarity matrix and a context embedding matrix; performing spectral decomposition and structure regulation on the edge weight weight matrix to obtain an edge weight embedding matrix, performing nonlinear cross mapping on the node similarity matrix, the context embedding matrix, and the edge weight embedding matrix to obtain a graph embedding matrix; performing node stability analysis on the graph embedding matrix to obtain a node cohesion index and a structure deviation, performing convolution attention aggregation on the node cohesion index and the structure deviation to obtain a comprehensive node coupling coefficient, and aggregating and reconstructing the graph embedding matrix and the comprehensive node coupling coefficient to obtain a node scene adaptation factor set.

[0009] As a preferred solution, the step of performing convolution attention aggregation on the node cohesion index and the structure deviation to obtain a comprehensive node coupling coefficient, and aggregating and reconstructing the graph embedding matrix and the comprehensive node coupling coefficient to obtain a node scene adaptation factor set comprises: performing multi-scale convolution on the node cohesion index and the structure deviation to obtain a plurality of node coupling coefficient candidate values, weighting and synthesizing the plurality of node coupling coefficient candidate values by using a convolution attention mechanism to obtain a comprehensive node coupling coefficient; adaptively weighting and aggregating the graph embedding matrix and the comprehensive node coupling coefficient to obtain a graph reconstruction matrix, and performing fine-grained deconstruction on the graph reconstruction matrix to extract a scene adaptation factor of each node to construct a node scene adaptation factor set.

[0010] As a preferred solution, the step of performing sequence clustering on the node scene adaptation factor set to generate a multi-level adaptation graph, and determining an adaptation scene cluster to which the current node to be adapted belongs based on the multi-level adaptation graph, comprises: performing time segmentation and trend extraction on the node scene adaptation factor set to obtain a node response trend sequence and a node steady-state feature sequence, performing clustering similarity compression on the node response trend sequence to obtain a response compressed representation vector, and performing aggregation mapping enhancement on the node steady-state feature sequence to obtain a steady-state feature representation matrix; performing structural dual-channel clustering on the response compressed representation vector and the steady-state feature representation matrix to obtain a candidate node clustering group, generating a node consistency matrix and an adaptation scene boundary matrix based on the candidate node clustering group; performing scene cluster identification on the adaptation scene boundary matrix and the node consistency matrix to obtain a multi-level adaptation graph, and performing hierarchical structure reconstruction on the multi-level adaptation graph to extract a cluster level center vector and an intra-cluster coupling relationship graph; and performing bidirectional matching on the current node to be adapted and the cluster level center vector and the intra-cluster coupling relationship graph to determine the adaptation scene cluster to which the current node to be adapted belongs.

[0011] As a preferred solution, the step of performing structural dual-channel clustering on the response compressed representation vector and the steady-state feature representation matrix to obtain a candidate node clustering group, and generating a node consistency matrix and an adaptation scene boundary matrix based on the candidate node clustering group, comprises: performing principal component analysis dimension reduction processing on the response compressed representation vector to obtain a variance feature vector, applying a density-based spatial clustering algorithm to perform clustering processing on the steady-state feature representation matrix to extract the density similarity between nodes, and obtaining a clustering result; fusing the variance feature vector and the clustering result to obtain a candidate node clustering group, calculating the similarity between each subset node in the candidate node clustering group using cosine similarity to obtain a graph similarity matrix between nodes; applying a graph matching algorithm to perform structural alignment verification on all the graph similarity matrices to obtain a structural alignment verification result; performing consistency verification on the adaptation relationship of all nodes based on the structural alignment verification result to obtain a node consistency matrix, and generating an adaptation scene boundary matrix using the node consistency matrix and the graph similarity matrix.

[0012] Preferably, the step of dynamically adjusting the charging scheduling strategy and parameter configuration scheme of the current node to be adapted according to the center feature vector of the adaptation scenario cluster comprises: performing difference calculation on the center feature vector of the adaptation scenario cluster and the environmental response vector of the current node to be adapted to obtain an environmental difference degree vector, and performing coupling fitting on the center feature vector of the adaptation scenario cluster and the behavior response vector of the current node to be adapted to obtain a behavior fit degree parameter; performing collaborative regression analysis on the environmental difference degree vector and the behavior fit degree parameter to obtain a scheduling strategy offset matrix and a parameter configuration offset vector, performing scheduling channel reconstruction and time slice redistribution on the scheduling strategy offset matrix to obtain an initial scheduling parameter set and a backup strategy path set, and performing distributed compression processing on the parameter configuration offset vector to obtain a charging power adjustment factor and a time frequency adjustment template; constructing a first scheduling scheme by using the initial scheduling parameter set and the charging power adjustment factor, constructing a second scheduling scheme by using the backup strategy path set and the time frequency adjustment template, and performing fusion judgment and matching screening on the first scheduling scheme and the second scheduling scheme to obtain a main scheduling scheme and a buffer backup scheme; and dynamically adjusting the charging scheduling strategy and parameter configuration scheme of the current node to be adapted by using the main scheduling scheme and the buffer backup scheme, wherein the main scheduling scheme is used for the current node execution process, and the buffer backup scheme is used for preloading of a strategy to be switched.

[0013] The application further provides a multi-scenario adaptation system of a charging network, comprising: an acquisition module, configured to acquire environmental parameter sequences and behavior parameter sequences of a plurality of charging nodes in a target area, and perform feature extraction on the environmental parameter sequences and the behavior parameter sequences to obtain environmental response vectors and behavior response vectors; an analysis module, configured to construct a composite adaptation graph by using the environmental response vectors and the behavior response vectors, perform multi-dimensional feature embedding on the composite adaptation graph to obtain a graph embedding matrix, and perform stability analysis on the graph embedding matrix to obtain a node scenario adaptation factor set; a clustering module, configured to perform sequence clustering on the node scenario adaptation factor set to generate a multi-level adaptation graph, and determine an adaptation scenario cluster to which a current node to be adapted belongs based on the multi-level adaptation graph, wherein the current node to be adapted refers to a node with a specific geographic location, user demand characteristics and historical charging data; and an adjustment module, configured to dynamically adjust a charging scheduling strategy and a parameter configuration scheme of the current node to be adapted according to a center feature vector of the adaptation scenario cluster.

[0014] Compared with the prior art, the application has the following beneficial effects: high scheduling efficiency and high flexibility. By performing feature extraction on the environmental parameter sequence and the behavior parameter sequence of multiple charging nodes in a target area, combining with the composite adaptive atlas construction of the environmental response vector and the behavior response vector, adaptive scheduling of the charging nodes in different scenarios is realized; through stability analysis of the atlas embedding matrix and clustering analysis of the node scene adaptation factor set, different scenarios can be accurately matched, and dynamic scheduling optimization is realized; finally, through adaptive scene cluster identification based on the node feature vector and charging strategy adjustment, not only the efficiency of the charging network is improved, but also the problems of low prediction and scheduling efficiency, poor flexibility and failure to fully utilize charging resources in the face of complex adaptive requirements in the process of scene adaptation in the traditional charging network are avoided. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0016] The structures, proportions, sizes, etc. shown in the drawings of the present specification are only used to cooperate with the content disclosed in the present specification, to enable those skilled in the art to understand and read, and are not used to limit the conditions that the present application can be implemented, so they do not have technical significance. Any modification of structure, change of proportion relationship or adjustment of size, which does not affect the effects that the present application can produce and the purposes that the present application can achieve, should still fall within the scope of the technical content disclosed by the present application.

[0017] Figure 1 is a flowchart of a multi-scene adaptation method of a charging network provided by the embodiment of the present application; Figure 2 is a structural schematic block diagram of a multi-scene adaptation system of a charging network provided by the embodiment of the present application.

[0018] Explanation of reference numerals: 10, multi-scene adaptation system of a charging network; 11, acquisition module; 12, analysis module; 13, clustering module; 14, adjustment module. DETAILED DESCRIPTION

[0019] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only some of the embodiments of the present application, but not all of them. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of the present application.

[0020] The flowcharts shown in the drawings are only illustrative, and do not necessarily include all contents and operations / steps, nor are they necessarily executed in the described order. For example, some operations / steps can be further decomposed, combined or partially merged, so that the actual execution order can be changed according to actual conditions.

[0021] It should also be understood that the terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification and the appended claims of the present application, the singular forms "a", "an" and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0022] It should be further understood that the term "and / or" used in the specification and the appended claims of the present application means any combination of one or more of the associated listed terms and all possible combinations thereof, and includes these combinations.

[0023] The technical solutions of the present application will be further described below in conjunction with the drawings and through specific embodiments.

[0024] Embodiment 1: As Figure 1 shown, the present application provides a multi-scene adaptation method of a charging network, including steps S100 to S400.

[0025] Step S100, acquiring the environmental parameter sequence and the behavior parameter sequence of a plurality of charging nodes in a target area, and performing feature extraction on the environmental parameter sequence and the behavior parameter sequence to obtain an environmental response vector and a behavior response vector.

[0026] In this step, the environmental parameter sequence and the behavior parameter sequence of a plurality of charging nodes in the target area are collected in real time through sensors and communication devices deployed at the charging nodes. Specifically, the environmental parameter sequence includes data such as the terrain type, climate condition, temperature fluctuation, etc. of the location where the charging node is located, and the behavior parameter sequence includes information such as user arrival frequency, stay time, charging power demand change, etc.

[0027] For example, in urban areas, the environmental parameters include geographical location (such as city center, highway surrounding), climate conditions (such as temperature, humidity, precipitation, etc.), and diurnal temperature difference; the behavioral parameters include charging request frequency during peak hours, average charging power demand, etc.

[0028] Step S200, constructing a composite adaptation graph using the environmental response vector and the behavioral response vector, performing multi-dimensional feature embedding on the composite adaptation graph to obtain a graph embedding matrix, and performing stability analysis on the graph embedding matrix to obtain a node scene adaptation factor set.

[0029] In this step, the environmental response vector and the behavioral response vector obtained through feature extraction are fused together to construct a composite adaptation graph. Specifically, the dimensions and data forms of the environmental response vector and the behavioral response vector are embedded through a specific algorithm (such as principal component analysis) to ensure that the graph can accurately reflect the adaptation characteristics of the nodes.

[0030] For example, by merging the environmental response vector and the behavioral response vector of each charging node in the same multi-dimensional space to form a composite adaptation graph, and then performing stability analysis on the graph, the adaptation factors of each node under different scenarios are calculated.

[0031] Step S300, performing sequence clustering on the node scene adaptation factor set to generate a multi-level adaptation graph, and determining the adaptation scene cluster to which the current to-be-adapted node belongs based on the multi-level adaptation graph, wherein the current to-be-adapted node refers to a node with a specific geographical location, user demand characteristics, and historical charging data.

[0032] In this step, the node scene adaptation factor set is subjected to sequence clustering, and a clustering algorithm (such as K-means, DBSCAN, etc.) is used to group the nodes. Specifically, the adaptation factor data of all charging nodes is clustered according to the time sequence, and the scene characteristics represented by each class of adaptation factor is identified.

[0033] For example, nodes can be divided into different adaptation scene clusters according to their surrounding environment (such as busy city center and quiet suburb) and user behavior (such as charging demand during peak hours and off-peak hours). Based on the clustering results, the system can identify the adaptation scene cluster of the current to-be-adapted node, thereby providing the most suitable scene selection for subsequent scheduling decisions.

[0034] Step S400, dynamically adjusting the charging scheduling strategy and parameter configuration scheme of the current to-be-adapted node according to the center feature vector of the adaptation scene cluster.

[0035] In this step, according to the adaptation scenario cluster obtained from the multi-level adaptation graph, the system will dynamically adjust the charging scheduling strategy and parameter configuration scheme of the to-be-adapted node according to the center feature vector of each cluster. Specifically, first, the center feature vector of the adaptation scenario cluster is determined, and then according to the feature information of the vector, the charging power, charging period, load of the charging station and other parameters of the to-be-adapted node are adjusted.

[0036] For example, in a high-demand scenario (such as a downtown peak period), the system will adjust the power output of the charging node to increase the charging capacity; while in a low-demand scenario (such as a suburban night period), the power output is reduced to reduce resource waste.

[0037] In this embodiment, the environmental parameter sequence and the behavior parameter sequence of a plurality of charging nodes in a target area are obtained. Then, the environmental response vector and the behavior response vector are obtained by feature extraction on the environmental parameter sequence and the behavior parameter sequence. Based on the environmental response vector and the behavior response vector, a composite adaptation graph is constructed. Subsequently, multi-dimensional feature embedding is performed on the composite adaptation graph to generate a graph embedding matrix, and stability analysis is performed on the matrix to obtain a set of node scenario adaptation factors. Then, sequence clustering is performed on all the node scenario adaptation factor sets to generate a multi-level adaptation graph. Based on the multi-level adaptation graph, the adaptation scenario cluster to which the current to-be-adapted node belongs is determined, wherein the current to-be-adapted node has a specific geographic location, user demand characteristics and historical charging data. Finally, the charging scheduling strategy and parameter configuration scheme of the current to-be-adapted node are dynamically adjusted according to the center feature vector of the adaptation scenario cluster. By constructing a composite adaptation graph and performing multi-dimensional feature embedding, the adaptability of the node in different scenarios can be better reflected, and the self-adaptation ability of the charging network in actual application can be improved. Through sequence clustering of the node scenario adaptation factors, the multi-level adaptation graph generated is helpful to distinguish different adaptation scenarios, and the charging scheduling strategy and parameter configuration scheme are dynamically adjusted according to the center feature vector of the scenario cluster, avoiding the low efficiency problem caused by static scheduling, so as to realize efficient and flexible scheduling of the charging network in different geographic locations and different user demands, greatly improving the utilization rate of charging resources, and solving the problem that the prediction and scheduling efficiency of charging demand is low and the flexibility is poor when facing complex adaptation demand, and the charging resources are not fully utilized.

[0038] Embodiment 2: In step S100, the environmental parameter sequence and the behavior parameter sequence of a plurality of charging nodes in a target area are collected by a pre-set sensor and a communication module.

[0039] The environmental and behavioral data of the charging node are collected through the sensors and the communication module. Specifically, the system installs multiple sensors on each charging node, including temperature, humidity, air pressure, light intensity, wind speed, and other environmental parameter sensors, as well as devices for recording user behavior (such as charging gun connection records, charging duration, power demand, etc.). These sensors collect environmental parameter sequences and behavioral parameter sequences of multiple charging nodes in the target area in real time and transmit the data to the central data platform for analysis and processing through the communication module.

[0040] For example, at a charging station in a certain urban area, the temperature sensor records that the temperature is as high as 30°C at certain times, and the humidity sensor monitors that the humidity changes sharply at certain times; at the same time, the behavioral parameter sensor in the charging station records the charging demand of each user, especially during the morning and evening peak hours, the charging request density is larger and the charging time is longer.

[0041] The environmental parameter sequences are processed by periodic segmentation and local fluctuation calculation to obtain the node terrain variation index, climate cycle pattern, and temperature difference distribution coefficient. The behavioral parameter sequences are processed by window weighting and abnormal point elimination to obtain the user stay time dynamic weight, charging request density matrix, and rhythm variation curve.

[0042] The environmental parameters are processed by periodic segmentation and local fluctuation analysis. Specifically, the system performs periodic segmentation (such as hourly, daily, or monthly segmentation) on the environmental parameter sequences, and performs local fluctuation calculation on the environmental data (such as temperature, humidity, light, etc.) in each segment to obtain the node terrain variation index, climate cycle pattern, and temperature difference distribution coefficient. Environmental fluctuation analysis can reveal the change characteristics of temperature, climate, and other factors, providing a basis for subsequent scheduling decisions.

[0043] For example, in a certain urban center area, the temperature changes sharply, and the system obtains a high temperature difference distribution coefficient through local fluctuation calculation, indicating that the climate environment fluctuation in this area is large, and additional consideration of environmental factors on charging load is needed. The behavioral parameter sequences are processed by window weighting algorithm to remove abnormal data and calculate the user stay time dynamic weight, charging request density matrix, and rhythm variation curve, which can more accurately reflect the user's charging demand and behavior rules.

[0044] The node terrain variation index and climate cycle pattern are combined and encoded to obtain the first environmental representation vector, the temperature difference distribution coefficient and the user stay time dynamic weight are normalized and fused to obtain the environmental behavior cross factor, and the charging request density matrix and the rhythm variation curve are convolved and fitted to obtain the user behavior spectrum vector.

[0045] The system combines and encodes the node terrain variation index and the climate cycle pattern to obtain a first environment representation vector, so as to quantify the geographical environment and climate change condition where the node is located. Meanwhile, the system normalizes and fuses the temperature difference distribution coefficient and the dynamic weight of user stay time to obtain an environment-behavior cross factor representing the combination characteristics of the environment and the behavior; then, the system combines the charging request density matrix and the rhythm variation curve through a convolution fitting algorithm to generate a user behavior spectrum vector reflecting the periodic characteristics of the user charging demand.

[0046] For example, in summer, the temperature of a certain charging station changes greatly, and the system obtains a higher terrain variation index and climate cycle pattern through combination encoding, indicating that the charging environment of this area is special. The user stays for a long time during the peak period, and the system combines the temperature difference and the stay time to obtain a higher environment-behavior cross factor, which helps to understand the behavior characteristics of the user in different environments.

[0047] The first environment representation vector and the environment-behavior cross factor are spliced to obtain an environment response vector, and the environment-behavior cross factor and the user behavior spectrum vector are combined to obtain a behavior response vector.

[0048] The system splices the first environment representation vector and the environment-behavior cross factor to generate an environment response vector, combines the environmental factors and the user behavior characteristics to obtain the environmental adaptability of the node; meanwhile, the system combines the environment-behavior cross factor and the user behavior spectrum vector to obtain a behavior response vector, representing the adaptability of the charging node under different time periods and different charging demands.

[0049] For example, in a cold area, the temperature is low, and the system splices the features of the terrain variation index and the climate pattern to obtain an environment response vector; the user stays for a short time and has less charging demand, and the system combines the environment-behavior cross factor and the user behavior spectrum vector to obtain a behavior response vector, further reflecting the user behavior characteristics of the charging node in this area.

[0050] In step S200, the node environment graph is constructed based on the environment response vector, the node behavior graph is constructed based on the behavior response vector, the node environment graph and the node behavior graph are fused at the graph level, and a composite adaptation graph is generated.

[0051] The system constructs a node environment graph based on the environmental response vector of each charging node, and each node in the graph contains the environmental characteristics (such as air temperature, humidity, geographical location, etc.) of the charging node. At the same time, based on the behavior response vector, the system constructs a node behavior graph, and each node represents the user behavior characteristics (such as arrival frequency, stay time, etc.) of the charging node. Then, the system merges the two graphs through graph-level joint fusion technology to form a composite adaptation graph, and the fused graph can comprehensively reflect the adaptability of the charging node under different environments and user behaviors.

[0052] For example, in a city area, the nodes in the node environment graph represent a climate section (such as high temperature and high humidity environment) where a charging station is located, and the node behavior graph includes the charging demand density of the charging station during peak hours. By jointly fusing the two graphs, the system can consider the influence of environmental factors and user behavior on charging demand, and generate a composite adaptation graph.

[0053] The node relationship tensor and edge weight weight matrix are obtained by nested multi-view feature extraction on the composite adaptation graph, and the node similarity matrix and context embedding matrix are obtained by tensor compression decomposition on the node relationship tensor.

[0054] Through multi-view feature extraction and tensor decomposition; specifically, the generated composite adaptation graph is subjected to nested multi-view feature extraction, and various feature extraction methods (such as local feature extraction and global feature extraction) are used to extract important information from the graph. Through these extractions, a node relationship tensor is generated, representing the mutual relationship between charging nodes. At the same time, the system also constructs an edge weight matrix, which is used to represent the strength of the connection between nodes. Then, the node relationship tensor is subjected to tensor compression decomposition to reduce computational complexity, and a node similarity matrix and a context embedding matrix are obtained, which are used to measure the similarity between nodes and the embedding features of nodes in the context, respectively.

[0055] For example, during high demand periods, the charging behavior (such as peak power demand) of some charging nodes has similarity with other nodes, and the similarity matrix obtained by the system through tensor decomposition can cluster these nodes into similar groups. At the same time, the node context embedding matrix helps the system identify the performance of different nodes in different charging scenarios.

[0056] The edge weight embedding matrix is obtained by spectral decomposition and structure regulation of the edge weight weight matrix, and the node similarity matrix, context embedding matrix, and edge weight embedding matrix are subjected to nonlinear cross mapping to obtain the graph embedding matrix.

[0057] By spectral decomposition and nonlinear cross mapping; specifically, the spectral decomposition is performed on the edge weight matrix to obtain the importance of each node in the graph spectrum, and the allocation of the edge weight is further optimized through structure regulation to ensure that the weight of the important node is higher. Then, the system performs nonlinear cross mapping on the node similarity matrix, the context embedding matrix and the edge weight embedding matrix to generate a graph spectrum embedding matrix, which can comprehensively reflect the adaptability and importance of the charging node in multiple dimensions.

[0058] For example, during peak hours, some charging station nodes are assigned higher weights because these nodes have stronger adaptability and response capability to charging demand, and the system optimizes the weights of these nodes through spectral decomposition to ensure that the charging scheduling prioritizes processing these high-demand nodes.

[0059] The node stability analysis is performed on the graph spectrum embedding matrix to obtain a node cohesion index and a structure deviation, the node cohesion index and the structure deviation are convolved and attention-aggregated to obtain a comprehensive node coupling coefficient, and the graph spectrum embedding matrix and the comprehensive node coupling coefficient are aggregated and reconstructed to obtain a node scene adaptation factor set.

[0060] By node stability analysis and convolution aggregation; specifically, the system performs node stability analysis on the graph spectrum embedding matrix, calculates the cohesion index (indicating the similarity of the node to its adjacent nodes) and the structure deviation (indicating the inconsistency of the structure between nodes) of each node in the composite adaptation graph to obtain two key indicators. Then, the node cohesion index and the structure deviation are aggregated using a convolution attention mechanism to obtain a comprehensive node coupling coefficient, and the relationship between the graph spectrum embedding matrix and the comprehensive node coupling coefficient is further analyzed to reconstruct the graph and generate a node scene adaptation factor set.

[0061] For example, for a charging node with high cohesion and low structure deviation, the system assigns a higher coupling coefficient to the node, indicating that the node has better adaptability in the charging network and is suitable for undertaking high-load charging tasks.

[0062] The steps of convolving and attention-aggregating the node cohesion index and the structure deviation to obtain a comprehensive node coupling coefficient, and aggregating and reconstructing the graph spectrum embedding matrix and the comprehensive node coupling coefficient to obtain a node scene adaptation factor set include: The node cohesion index and the structure deviation are multi-scale convolved to obtain a plurality of node coupling coefficient candidate values, and the convolution attention mechanism is used to weight and synthesize the plurality of node coupling coefficient candidate values to obtain a comprehensive node coupling coefficient.

[0063] By convolution analysis on the node cohesion index and the structure deviation, specifically, the system respectively performs convolution calculation on the node cohesion index and the structure deviation of each node, adopts multi-scale convolution kernels (such as 3x3, 5x5, 7x7, etc.) to perform feature extraction on the data, and ensures that the local and global features of different nodes are effectively captured. Multi-scale convolution can identify the similarity and difference between nodes in different scenarios. Subsequently, the system calculates the coupling candidate value of the node through multiple perspective weight adjustment, and obtains multiple candidate value matrices. Then, the convolution attention mechanism is used to weight and synthesize these candidate values, the optimal coupling value is identified through the self-attention mechanism, and is fused, and finally the comprehensive node coupling coefficient is obtained.

[0064] For example, in a high demand scenario, some charging nodes have high cohesion and low structure deviation, the system captures the global characteristics of these nodes through large convolution kernels, and analyzes the local features of the nodes through small convolution kernels, finally calculates multiple candidate value matrices, and assigns weights to the optimal coupling value through the attention mechanism to generate the comprehensive node coupling coefficient.

[0065] The graph embedding matrix and the comprehensive node coupling coefficient are adaptively weighted and aggregated to obtain a graph reconstruction matrix, and the graph reconstruction matrix is deconstructed in fine granularity to extract the scene adaptation factor of each node to construct a node scene adaptation factor set.

[0066] Through the fusion of the graph and the coupling coefficient and the extraction of the scene factor, specifically, the system first adaptively weights and aggregates the graph embedding matrix and the comprehensive node coupling coefficient, adjusts and fuses the graph embedding matrix by taking the coupling coefficient as the weight, and generates a graph reconstruction matrix. The graph reconstruction matrix reflects the comprehensive environmental and behavioral adaptability of the node. Subsequently, the system deconstructs the graph reconstruction matrix in fine granularity, calculates the scene feature index of each node, extracts the scene adaptation factor to which each node belongs, and finally generates a node scene adaptation factor set. Each factor set can indicate the adaptation characteristics of the charging node in a specific scenario, providing detailed basis for subsequent charging scheduling optimization.

[0067] For example, for a node with high behavioral adaptability and low environmental adaptability, the system identifies these features through the graph reconstruction matrix, and extracts the behavioral scene factor and the environmental scene factor of the node in the deconstruction process, finally generates the scene adaptation factor set of the node, and determines that the node is suitable for the scenario with high charging request density and large environmental fluctuation.

[0068] In step S300, the node scene adaptation factor set is time segmented and trend extracted to obtain a node response trend sequence and a node steady-state feature sequence, the node response trend sequence is clustered and similarity compressed to obtain a response compressed representation vector, and the node steady-state feature sequence is aggregated and mapped to enhance to obtain a steady-state feature representation matrix.

[0069] Through time segmentation and trend extraction; specifically, the system first segments the data according to the time axis for the node scene adaptation factor set, and the segmentation period is set to be hourly, daily, or weekly according to actual needs. After segmentation, the system analyzes the trend of each segment of data, respectively calculates the behavior dynamics and environmental stability of the node in the segment, and generates a node response trend sequence and a node steady-state feature sequence. The response trend sequence uses a sliding window method to extract the behavior change law of the node in different periods, and the steady-state feature sequence extracts environmental and behavior characteristic indicators of the node in a stable working state through statistical analysis. To further compress the data in the response trend sequence, the system applies a clustering similarity algorithm (such as K-means clustering) to the behavior change characteristic sequence of the node to reduce redundant data and retain main features, forming a response compressed representation vector. At the same time, a multi-channel feature aggregation algorithm (such as the bag-of-features algorithm) is used for the steady-state feature sequence to enhance its expression ability and perform multi-angle mapping, thereby obtaining a steady-state feature representation matrix.

[0070] For example, during the morning rush hour on weekdays, the user behavior of some nodes changes significantly, such as a sharp increase in charging request density and an extension of stay time. The system records these trends through time segmentation and generates a response trend sequence, and generates a steady-state feature sequence for the daily performance of the node (such as average charging efficiency), compresses the short-term changes (such as behavior anomalies before and after the peak) in the response trend, and maps the steady-state features (such as different charging efficiencies during the day and at night) to form a response compressed representation vector and a steady-state feature representation matrix.

[0071] Structural dual-channel clustering of the response compressed representation vector and the steady-state feature representation matrix obtains a candidate node clustering group, and a node consistency matrix and an adaptation scene boundary matrix are generated based on the candidate node clustering group.

[0072] The system performs structural dual-channel clustering on the response compressed representation vector and the steady-state feature representation matrix to obtain candidate node clustering groups. The system generates a node consistency matrix and an adaptive scenario boundary matrix based on the candidate node clustering groups. The steps of generating the node consistency matrix and the adaptive scenario boundary matrix based on the candidate node clustering groups include:

[0073] For example, in a certain region, the dynamic response characteristics of a node group are characterized by a sudden increase in charging demand during the morning peak, and the steady-state characteristics include stable operation in a low-temperature environment. The system classifies such nodes into a high-demand scenario through dual-channel clustering, and generates the stability relationship within the node group through the consistency matrix, while obtaining the related scenario boundary (such as the characteristics of the connecting nodes from the city center to the suburbs).

[0074] The system performs structural dual-channel clustering on the response compressed representation vector and the steady-state feature representation matrix to obtain candidate node clustering groups. The system generates a node consistency matrix and an adaptive scenario boundary matrix based on the candidate node clustering groups. The steps of generating the node consistency matrix and the adaptive scenario boundary matrix based on the candidate node clustering groups include: The system performs principal component analysis on the response compressed representation vector to obtain a variance feature vector, and applies a density-based spatial clustering algorithm to the steady-state feature representation matrix to extract the density similarity between nodes and obtain a clustering result.

[0075] The system performs principal component analysis (PCA) on the response compressed representation vector to extract key dimensions in the feature vector, removes noise, and retains high-variance parts to generate a variance feature vector. Then, the system applies a density-based spatial clustering algorithm (such as DBSCAN) to group nodes based on the distance properties of the steady-state feature representation matrix, thereby extracting groups of nodes with density correlations and generating a clustering result.

[0076] For example, some nodes exhibit sustained adaptability in a stable environment, but their behavior dynamics exhibit sparsity and high volatility. The system extracts high-variance factors through principal component analysis, and captures group characteristics (such as long-term high charging efficiency) in the steady-state feature space through density clustering.

[0077] The variance eigenvector is fused with the clustering result to obtain a candidate node clustering group, and a cosine similarity is used to calculate the similarity between each subset node in the candidate node clustering group to obtain a graph similarity matrix between nodes.

[0078] Through collaborative clustering and similarity analysis, specifically, the system uses an exact collaborative clustering algorithm to cross-dimensionally fuse the variance eigenvector and the density clustering result, and jointly matches the combination of the two characteristics of the node in response to dynamics and steady-state adaptation. Then, vector similarity calculation (such as cosine similarity) is applied to the generated candidate node clustering group to calculate the similarity coefficient between each subset node and generate a graph similarity matrix.

[0079] For example, in the morning peak and night valley scenarios, part of the charging nodes have completely different dynamic characteristics, but consistent steady-state adaptation performance. The system fuses the two-dimensional characteristics through collaborative clustering and identifies the group relationship, and improves the accuracy of the relationship between nodes through cosine similarity.

[0080] A structure alignment verification result is obtained by applying a graph matching algorithm to all graph similarity matrices.

[0081] Verification is performed through a graph matching algorithm; specifically, the system applies a graph matching algorithm (such as the Hungarian algorithm or the maximum flow graph algorithm) to the graph similarity matrix within each clustering group for node-by-node operation, optimizes the structure relationship within the group, and verifies the connectivity consistency between nodes, and finally generates a structure alignment verification result.

[0082] For example, for a group of urban center nodes, the system discovers the high consistency of the connection mode by calculating the optimal matching path, while eliminating a small number of incorrect connection edges, and further optimizes the graph through structure verification.

[0083] Based on the structure alignment verification result, the consistency of the adaptation relationship of all nodes is verified to obtain a node consistency matrix, and the adaptation scenario boundary matrix is generated using the node consistency matrix and the graph similarity matrix.

[0084] Through consistency relationship and boundary extraction; specifically, the node consistency matrix is generated according to the verification result, indicating the intra-group and cross-group stability relationship of each node. Then, the adaptation scenario boundary matrix is generated by analyzing the boundary between scenarios, indicating the division of different scenario characteristics and connection strength.

[0085] For example, in a certain region, the system identifies that the strength of the boundary connection between the high-demand scenario and the low-demand scenario decreases (such as a decrease in user demand after cooling), and finally generates a matrix describing the boundary properties, providing a basis for subsequent adaptation optimization.

[0086] The scene cluster identification is performed on the adaptive scene boundary matrix and the node consistency matrix to obtain a multi-level adaptive graph, and the multi-level adaptive graph is reconstructed to extract a cluster level center vector and a cluster internal coupling relationship graph.

[0087] The scene cluster identification and graph reconstruction are performed. Specifically, the system analyzes the scene boundary and the consistency matrix, divides the multi-level adaptive scene cluster, optimizes the generated graph level by level, extracts the global mean vector (center vector) of the cluster level and the connection relationship graph in the cluster, and ensures the completeness of the adaptive system structure of the scene cluster.

[0088] For example, for two overlapping level scene clusters (highway and suburban connection), the system optimizes the level division to ensure the completeness of the coupling between nodes and generates a key adaptive vector.

[0089] The current adaptive node is bidirectionally matched with the cluster level center vector and the cluster internal coupling relationship graph to determine the adaptive scene cluster to which the current adaptive node belongs.

[0090] The bidirectional matching technology is used. Specifically, the features of the adaptive node are statically matched with the mean value attributes of the cluster level center, and the dynamic relationship analysis is performed in combination with the internal relationship graph to determine the adaptive relationship of the node in the scene cluster and the category to which the node belongs.

[0091] For example, for a city center adaptive node, the system uses the mean value to match the adaptive degree and analyzes the connection behavior through the internal coupling relationship to finally accurately classify it into the city demand scene cluster.

[0092] In step S400, the center feature vector of the adaptive scene cluster and the environment response vector of the current adaptive node are difference calculated to obtain an environment difference vector, and the center feature vector of the adaptive scene cluster and the behavior response vector of the current adaptive node are coupled to fit to obtain a behavior fit degree parameter.

[0093] The difference calculation and coupling fitting are performed. Specifically, the system first obtains the center feature vector of the adaptive scene cluster and the environment response vector of the current adaptive node, compares the dimensions of the two to calculate the difference in the environment features. The difference calculation uses Euclidean distance, Mahalanobis distance or dynamic time warping (DTW) algorithm to generate an environment difference vector for quantifying the environment deviation degree between the current node and the scene cluster center. At the same time, the system obtains the center feature vector of the adaptive scene cluster and the behavior response vector of the current adaptive node, calculates the adaptive degree between the two through a coupling fitting algorithm (such as least square fitting or deep learning fitting) to generate a behavior fit degree parameter. The behavior fit degree reflects the consistency of the node charging behavior and the scene center features.

[0094] For example, the current to-be-adapted node is located in the suburbs of a city, and its environment response vector shows that the temperature is low and the humidity is high, while the center feature vector of the city center scenario cluster shows that the temperature is high and the humidity is low. The system generates an environment difference vector using a difference calculation algorithm, indicating that the to-be-adapted node environment deviates greatly from the center of the scenario. At the same time, the behavior response vector of the to-be-adapted node (such as low demand characteristics) has some overlap with the city center scenario cluster, and the behavior fit degree is calculated by coupling fitting, indicating that the behavior characteristics of the node partially meet the requirements of the scenario.

[0095] The environment difference vector and the behavior fit degree parameter are subjected to collaborative regression analysis to obtain a scheduling strategy offset matrix and a parameter configuration offset vector. The scheduling strategy offset matrix is subjected to scheduling channel reconstruction and time slice redistribution to obtain an initial scheduling parameter set and a backup strategy path set. The parameter configuration offset vector is subjected to distributed compression processing to obtain a charging power adjustment factor and a time frequency adjustment template.

[0096] Through collaborative regression analysis and target calculation; specifically, the system takes the environment difference vector and the behavior fit degree parameter as a collaborative regression analysis-based model, combines linear regression or tree model regression to calculate the offset of the node characteristics, generates a scheduling strategy offset matrix to describe the differences of the node in the scheduling scheme. At the same time, a parameter configuration offset vector is generated to represent the adjustment requirements of the node charging parameters. The scheduling strategy offset matrix is further used for channel reconstruction, and an optimized channel is formed through priority weight adjustment between channels, and the predicted time slices are finely redistributed to obtain an initial scheduling parameter set and a backup strategy path set. The parameter configuration offset vector is subjected to distributed compression processing to extract a charging power adjustment factor and a time frequency adjustment template, quantifying the charging capacity requirements and scheduling time correction results.

[0097] For example, the environment response vector of the current to-be-adapted node deviates greatly from the city center scenario cluster, while its behavior fit degree parameter shows that the node still has partial adaptability. The system obtains a scheduling strategy offset matrix through collaborative regression analysis, in which the priority weight is low and the adaptation time slice is concentrated in the off-peak period. The system generates an offset configuration vector, and extracts a charging power adjustment factor (such as reducing the charging power requirement) and a time frequency adjustment template (such as reducing the scheduling frequency in the peak period) through distributed compression processing.

[0098] An initial scheduling parameter set and a charging power adjustment factor are used to construct a first scheduling scheme, a backup strategy path set and a time frequency adjustment template are used to construct a second scheduling scheme, and the first scheduling scheme and the second scheduling scheme are subjected to fusion judgment and matching screening to obtain a main scheduling scheme and a buffer backup scheme.

[0099] The construction and matching optimization of the scheduling scheme; specifically, the initial scheduling parameter set generates a first scheduling scheme according to the current node adaptation scene and offset characteristics, which mainly includes the selection of adaptive channels, the configuration of time period matching points, power optimization and behavior parameter correction values. The standby strategy path set and the time frequency adjustment template are used to construct a second scheduling scheme, which corresponds to the channel standby path and standby time period strategy of the providing node. The system fuses and selects the two scheduling schemes through a dynamic decision algorithm (such as AHP hierarchical comprehensive method or deep reinforcement learning algorithm) to generate the final main scheduling scheme and buffer standby scheme.

[0100] For example, for a node in a city suburb, charging adaptation in a low demand period is taken as the core of the first scheduling scheme, and the second scheme is generated by identifying the standby channel path through the time frequency template. The system finds through matching optimization that the standby scheme is suitable for efficient scheduling of the node's night emergency demand, so it generates the main scheme for balanced resource allocation and the standby scheme for emergency demand switching.

[0101] The main scheduling scheme and the buffer standby scheme are used to dynamically adjust the charging scheduling strategy and parameter configuration scheme of the current node to be adapted, wherein the main scheduling scheme is used for the current node execution process, and the buffer standby scheme is used for preloading of the strategy to be switched.

[0102] Through dynamic adjustment and preloading optimization; specifically, the system adjusts the running of the node to be adapted according to the main scheduling scheme, including dynamic configuration of output power, real-time correction of time period operation, etc., while loading the buffer standby scheme into the node scheduling plan to meet the emergency demand and scene changes. In the running process, the system monitors the behavior changes of the node in real time, switches the main scheme to the standby scheme through a multi-task scheduling algorithm, and ensures the efficiency and robustness of the charging scheduling.

[0103] For example, the current node serves non-peak user demand with low power through the main scheme, while the standby scheme preloading ensures that it can be switched at any time in the event of a peak request. In the case of real-time scene changes, the system can quickly adjust the current scheduling strategy through dynamic monitoring, ensuring that the charging demand is met and maximizing resource utilization.

[0104] In this embodiment, through real-time collection and processing of the environmental parameter sequence and the behavior parameter sequence of the plurality of charging nodes in the target area, the system realizes feature extraction of the environmental response vector and the behavior response vector of the charging nodes, and optimizes the scene adaptation effect in depth in combination with the difference and inertial steady-state characteristics. First, the system extracts the node response trend sequence and the steady-state characteristic sequence through time segmentation and trend extraction, and forms the response compressed representation vector and the steady-state characteristic representation matrix by using clustering similarity compression and multi-channel feature mapping enhancement. These key features are fused by a structural double-channel clustering algorithm to generate a candidate node clustering group, and a node consistency matrix and an adaptation scene boundary matrix are generated based on graph similarity analysis and matching verification between nodes, realizing scene division and characteristic correlation optimization of different charging nodes. In the scene level processing, the cluster level center vector and the intra-cluster coupling relationship graph are extracted by scene cluster identification and multi-level adaptation graph reconstruction and optimization, and the system successfully constructs a dynamic decision-making framework matching the nodes and the scenes. Finally, through difference calculation, coupling fitting, collaborative regression analysis and scheduling scheme construction, the system formulates a main scheduling scheme and a buffer standby scheme, and realizes dynamic adjustment of the charging strategy and optimization of the resources of the nodes. In this embodiment, the scheme used significantly improves the charging network adaptation efficiency, so that the system can realize adaptive decision-making based on environmental and behavioral responses, meet the diversified needs of charging nodes in multiple scenes and the flexible adjustment of scene switching, thereby effectively improving the accuracy, stability and user experience of the charging scheduling system.

[0105] Embodiment 3: As shown in Figure 2 The present application also provides a multi-scene adaptation system 10 of a charging network, comprising an acquisition module 11, an analysis module 12, a clustering module 13 and an adjustment module 14.

[0106] The acquisition module 11 is mainly used for acquiring the environmental parameter sequence and the behavior parameter sequence of a plurality of charging nodes in a target area, performing feature extraction on the environmental parameter sequence and the behavior parameter sequence to obtain an environmental response vector and a behavior response vector.

[0107] The analysis module 12 is mainly used for constructing a composite adaptation graph by using the environmental response vector and the behavior response vector, performing multi-dimensional feature embedding on the composite adaptation graph to obtain a graph embedding matrix, and performing stability analysis on the graph embedding matrix to obtain a node scene adaptation factor set.

[0108] The clustering module 13 is mainly used for performing sequence clustering on the node scene adaptation factor set to generate a multi-level adaptation graph, and determining an adaptation scene cluster to which a current to-be-adapted node belongs based on the multi-level adaptation graph, wherein the current to-be-adapted node refers to a node having a specific geographic location, user demand characteristics and historical charging data.

[0109] The adjusting module 14 is mainly used for dynamically adjusting the charging scheduling strategy and parameter configuration scheme of the current to-be-adapted node according to the center feature vector of the adaptation scene cluster.

[0110] In the embodiment, through the cooperative work of the acquiring module 11, the analyzing module 12, the clustering module 13 and the adjusting module 14, the system realizes the multi-scene adaptation function of the charging network. The acquiring module 11 is responsible for acquiring the environmental parameter sequence and the behavior parameter sequence of multiple charging nodes in the target area in real time, and generating the environmental response vector and the behavior response vector through feature extraction, ensuring the data availability and accuracy of the charging nodes under complex environment and dynamic demand changes. The analyzing module 12 further constructs a composite adaptation graph by using the environmental response vector and the behavior response vector, generates a graph embedding matrix and a node scene adaptation factor set through multi-dimensional feature embedding and graph stability analysis, ensuring the fine expression of the characteristics extraction and adaptation relationship of the node under the influence of multiple scenes. The clustering module 13 performs sequence clustering analysis on the scene adaptation factor set, accurately identifies the adaptation scene cluster to which the current to-be-adapted node belongs through the generation of a multi-level adaptation graph, and realizes the hierarchical optimization of scene division and the matching association of nodes and scenes. The adjusting module 14 performs difference calculation and behavior fitting on the response vector of the current to-be-adapted node according to the center feature vector of the adaptation scene cluster, and realizes the dynamic adjustment and optimization of the scheduling strategy and the parameter configuration based on the offset matrix and the scheme construction logic. In the embodiment, the cooperative work of the four modules enables the system to real-time perceive and analyze the influence of different geographical environments and user behaviors on the charging nodes, so as to dynamically adapt to different scenes, meet the changes of charging demand and the flexibility of scene switching, and significantly improve the scheduling accuracy and resource utilization rate of the charging network, providing efficient and reliable charging service support for users.

[0111] It should be noted that, for the convenience and brevity of description, the specific working processes of the above-described system and modules can refer to the corresponding processes in the foregoing embodiment 1, which will not be described herein.

[0112] The structures, proportions, sizes, etc. shown in the drawings of the present specification are only used to cooperate with the content disclosed in the present specification, to enable those skilled in the art to understand and read, and do not define the limiting conditions for the implementation of the present application, and therefore do not have substantial technical significance. Any modification of structure, change of proportion relationship or adjustment of size, which does not affect the effects and purposes that can be achieved by the present application, should still fall within the scope of the technical content disclosed by the present application.

[0113] The above-described embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those skilled in the art that the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced by equivalent replacements; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A multi-scenario adaptation method for a charging network, characterized in that: include: Obtaining environmental parameter sequences and behavioral parameter sequences of multiple charging nodes in a target area, performing feature extraction on the environmental parameter sequences and the behavioral parameter sequences, and obtaining environmental response vectors and behavioral response vectors; Constructing a complex matching graph using the environmental response vector and the behavioral response vector, performing multi-dimensional feature embedding on the complex matching graph to obtain a graph embedding matrix, performing stability analysis on the graph embedding matrix to obtain a set of node scenario adaptation factors; Performing sequential clustering on the node scenario adaptation factor set to generate a multi-level adaptation graph, and determining the adaptation scenario cluster to which the current node to be adapted belongs based on the multi-level adaptation graph, wherein the current node to be adapted refers to a node with a specific geographical location, user demand characteristics, and historical charging data; The charging scheduling strategy and parameter configuration scheme of the current node to be adapted are dynamically adjusted according to the central feature vector of the adaptation scenario cluster.

2. The multi-scenario adaptation method of the charging network according to claim 1, characterized in that: The step of obtaining environmental parameter sequences and behavioral parameter sequences of multiple charging nodes in the target area, performing feature extraction on the environmental parameter sequences and the behavioral parameter sequences, and obtaining environmental response vectors and behavioral response vectors includes: Collect environmental parameter sequences and behavioral parameter sequences of multiple charging nodes in the target area through preset sensors and communication modules; The environmental parameter sequence is subjected to periodic segmentation processing and local fluctuation calculation to obtain node terrain variation index, climate cycle pattern and temperature difference distribution coefficient; the behavioral parameter sequence is subjected to window weighting processing and outlier removal to obtain user stay time dynamic weight, charging request density matrix and rhythm change curve; The node terrain variation index and the climate cycle pattern are combined and encoded to obtain a first environment characterization vector, the temperature difference distribution coefficient is normalized and fused with the dynamic weight of the user's residence time to obtain an environmental behavior cross factor, and the charging request density matrix is ​​convoluted and fitted with the rhythm variation curve to obtain a user behavior spectrum vector; The first environment characterization vector and the environment behavior cross factor are feature concatenated to obtain an environment response vector, and the environment behavior cross factor and the user behavior spectrum vector are feature residual synthesized to obtain a behavior response vector.

3. The multi-scenario adaptation method of the charging network according to claim 1, characterized in that: The steps of constructing a complex matching graph using the environmental response vector and the behavioral response vector, performing multi-dimensional feature embedding on the complex matching graph to obtain a graph embedding matrix, performing stability analysis on the graph embedding matrix, and obtaining a set of node scenario adaptation factors include: Constructing a node environment graph based on the environment response vector, constructing a node behavior graph based on the behavior response vector, and performing graph-level joint fusion on the node environment graph and the node behavior graph to generate a composite matching graph; Performing nested multi-view feature extraction on the composite matching graph to obtain a node relationship tensor and an edge weight matrix, and performing tensor compression decomposition on the node relationship tensor to obtain a node similarity matrix and a context embedding matrix; Performing spectral decomposition and structural regulation on the edge weight matrix to obtain an edge weight embedding matrix, and performing nonlinear cross-mapping on the node similarity matrix, the context embedding matrix, and the edge weight embedding matrix to obtain a graph embedding matrix; A node stability analysis is performed on the graph embedding matrix to obtain a node cohesion index and a structural offset, the node cohesion index and the structural offset are subjected to convolutional attention aggregation to obtain a comprehensive node coupling coefficient, the graph embedding matrix and the comprehensive node coupling coefficient are aggregated and reconstructed to obtain a set of node scene adaptation factors.

4. The multi-scenario adaptation method of the charging network according to claim 3, characterized in that: The steps of performing convolutional attention aggregation on the node cohesion index and the structural offset to obtain a comprehensive node coupling coefficient, and aggregating and reconstructing the graph embedding matrix and the comprehensive node coupling coefficient to obtain a set of node scene adaptation factors include: Performing multi-scale convolution on the node cohesion index and the structural offset to obtain multiple node coupling coefficient candidate values, and performing weighted synthesis on the multiple node coupling coefficient candidate values ​​using a convolutional attention mechanism to obtain a comprehensive node coupling coefficient; The graph embedding matrix and the comprehensive node coupling coefficient are adaptively weighted and aggregated to obtain a graph reconstruction matrix. The graph reconstruction matrix is ​​fine-grainedly deconstructed to extract the scene adaptation factor of each node to construct a node scene adaptation factor set.

5. The multi-scenario adaptation method of the charging network according to claim 1, characterized in that: The step of sequentially clustering the node scene adaptation factor set to generate a multi-level adaptation graph, and determining the adaptation scene cluster to which the current node to be adapted belongs based on the multi-level adaptation graph includes: Performing time segmentation and trend extraction on the node scene adaptation factor set to obtain a node response trend sequence and a node steady-state feature sequence, performing clustering similarity compression on the node response trend sequence to obtain a response compression representation vector, and performing cluster mapping enhancement on the node steady-state feature sequence to obtain a steady-state feature representation matrix; Performing structural dual-channel clustering on the response compression representation vector and the steady-state feature representation matrix to obtain a candidate node cluster group, and generating a node consistency matrix and an adaptation scene boundary matrix based on the candidate node cluster group; Performing scene cluster identification on the adaptation scene boundary matrix and the node consistency matrix to obtain a multi-level adaptation map, reconstructing the hierarchical structure of the multi-level adaptation map, and extracting cluster-level center vectors and intra-cluster coupling relationship maps; The current node to be adapted is bidirectionally matched with the cluster-level center vector and the intra-cluster coupling relationship graph to determine the adaptation scenario cluster to which the current node to be adapted belongs.

6. The multi-scenario adaptation method of the charging network according to claim 5, characterized in that: The step of performing structural dual-channel clustering on the response compression representation vector and the steady-state feature representation matrix to obtain a candidate node cluster group, and generating a node consistency matrix and an adaptation scene boundary matrix based on the candidate node cluster group includes: Performing principal component analysis and dimensionality reduction processing on the response compression representation vector to obtain a variance eigenvector, applying a density-based spatial clustering algorithm to cluster the steady-state feature representation matrix, extracting density similarity between nodes, and obtaining a clustering result; The variance eigenvector is merged with the clustering result to obtain a candidate node cluster group, and the similarity between each subset node in the candidate node cluster group is calculated using cosine similarity to obtain a graph similarity matrix between nodes; Applying a graph matching algorithm to perform structural alignment verification on all of the graph similarity matrices to obtain a structural alignment verification result; Based on the structural alignment verification result, the adaptation relationship of all nodes is checked for consistency to obtain a node consistency matrix, and the node consistency matrix and the graph similarity matrix are used to generate an adaptation scene boundary matrix.

7. The multi-scenario adaptation method of the charging network according to claim 1, characterized in that: The step of dynamically adjusting the charging scheduling strategy and parameter configuration scheme of the current node to be adapted according to the central feature vector of the adaptation scenario cluster includes: Performing a difference calculation between the central feature vector of the adaptation scene cluster and the environmental response vector of the current node to be adapted to obtain an environmental difference vector, and performing coupling fitting between the central feature vector of the adaptation scene cluster and the behavioral response vector of the current node to be adapted to obtain a behavioral fit parameter; Performing collaborative regression analysis on the environmental difference vector and the behavior compatibility parameter to obtain a scheduling strategy offset matrix and a parameter configuration offset vector. Performing scheduling channel reconstruction and time slice reallocation on the scheduling strategy offset matrix to obtain an initial scheduling parameter set and a backup strategy path set. Performing distributed compression processing on the parameter configuration offset vector to obtain a charging power adjustment factor and a time frequency adjustment template. A first scheduling plan is constructed using the initial scheduling parameter set and the charging power adjustment factor, a second scheduling plan is constructed using the backup strategy path set and the time frequency adjustment template, and the first scheduling plan and the second scheduling plan are integrated, judged, matched, and screened to obtain a main scheduling plan and a buffer backup plan; The main scheduling scheme and the buffer standby scheme are used to dynamically adjust the charging scheduling strategy and parameter configuration scheme of the current node to be adapted, wherein the main scheduling scheme is used for the current node execution process, and the buffer standby scheme is used for preloading the strategy to be switched.

8. A multi-scenario adaptation system for a charging network, characterized in that: include: An acquisition module is used to acquire environmental parameter sequences and behavioral parameter sequences of multiple charging nodes in a target area, perform feature extraction on the environmental parameter sequences and the behavioral parameter sequences, and obtain an environmental response vector and a behavioral response vector; an analysis module, configured to construct a complex matching graph using the environmental response vector and the behavioral response vector, perform multi-dimensional feature embedding on the complex matching graph to obtain a graph embedding matrix, perform stability analysis on the graph embedding matrix, and obtain a set of node scenario adaptation factors; a clustering module, configured to sequentially cluster the node scenario adaptation factor set to generate a multi-level adaptation graph, and determine, based on the multi-level adaptation graph, the adaptation scenario cluster to which the current node to be adapted belongs, wherein the current node to be adapted refers to a node with a specific geographical location, user demand characteristics, and historical charging data; An adjustment module is used to dynamically adjust the charging scheduling strategy and parameter configuration scheme of the current node to be adapted according to the central feature vector of the adaptation scenario cluster.

Images