Iot-based charging pile intelligent operation management method and system

Abstract

The application relates to a charging pile intelligent operation management method and system based on the Internet of Things. The method comprises the following steps: constructing an Internet of Things topology for a charging pile node to obtain a node communication distance matrix and a network connectivity matrix; performing deep learning dimension reduction and node screening based on the network connectivity matrix to obtain a fault feature vector; inputting the fault feature vector and charging pile operation state data into a hybrid neural network for risk calculation to obtain a fault risk assessment value, wherein the hybrid neural network comprises a spatial perception subnetwork and a time sequence analysis subnetwork; performing normalization and dimension reduction mapping on charging pile power response data to obtain a time domain mapping matrix; performing particle swarm optimization calculation based on the time domain mapping matrix and the fault risk assessment value to obtain a charging power distribution scheme, and generating a control instruction set according to the charging power distribution scheme. The application realizes dynamic optimization of the charging power distribution scheme, and effectively balances charging efficiency and load balancing.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation technology for charging piles, and in particular to an intelligent operation management method and system for charging piles based on the Internet of Things. Background Technology

[0002] Traditional charging pile management methods mainly rely on manual monitoring and fixed rule scheduling, which are difficult to cope with increasingly complex charging demand fluctuations and equipment failures, and cannot fully realize the overall efficiency of charging pile clusters.

[0003] Currently, the Dynamic Safety Assessment (DSA) method for charging pile systems has significant shortcomings: on the one hand, existing data-driven smart charging pile DSA methods lack generalizable fault location features, resulting in low accuracy in assessing unlearned fault locations; on the other hand, traditional DSA methods often treat fault locations as fixed inputs, ignoring the impact of fault location uncertainty on system safety, thus questioning the reliability of the assessment results. Furthermore, in today's IoT-connected charging pile systems, the demand for flexible scheduling of smart charging piles has increased significantly, while existing dynamic optimal energy flow technologies still have considerable room for improvement in accuracy, efficiency, and convergence. Especially under the influence of uncertainties such as fluctuating charging demand and charging pile equipment failures, how to achieve coordinated control and intelligent scheduling of charging pile groups has become a critical technical challenge that urgently needs to be overcome. Summary of the Invention

[0004] The main objective of this invention is to provide a smart operation and management method and system for charging piles based on the Internet of Things. This invention achieves dynamic optimization of the charging power allocation scheme, effectively balancing charging efficiency and load balance.

[0005] To achieve the above objectives, this invention provides a smart operation and management method for charging piles based on the Internet of Things, comprising the following steps:

[0006] IoT topology construction is performed on charging pile nodes to obtain node communication distance matrix and network connectivity matrix;

[0007] Based on the network connectivity matrix, deep learning dimensionality reduction and node selection are performed to obtain the fault feature vector;

[0008] The fault feature vector and the charging pile operating status data are input into a hybrid neural network for risk calculation to obtain a fault risk assessment value. The hybrid neural network includes a spatial perception subnetwork and a time series analysis subnetwork.

[0009] The power response data of the charging pile is normalized and dimensionally reduced to obtain the time-domain mapping matrix;

[0010] Based on the time-domain mapping matrix and the fault risk assessment value, particle swarm optimization calculation is performed to obtain a charging power allocation scheme. A set of control commands is then generated according to the charging power allocation scheme, and the set of control commands includes the power and time parameters of each charging pile.

[0011] This invention also provides an IoT-based intelligent operation and management system for charging piles, comprising:

[0012] The module is used to construct the IoT topology of charging pile nodes, and obtain the node communication distance matrix and network connectivity matrix.

[0013] The filtering module is used to perform deep learning dimensionality reduction and node filtering based on the network connectivity matrix to obtain a fault feature vector.

[0014] The calculation module is used to input the fault feature vector and the charging pile operation status data into the hybrid neural network for risk calculation to obtain the fault risk assessment value. The hybrid neural network includes a spatial perception subnetwork and a time series analysis subnetwork.

[0015] The mapping module is used to normalize and reduce the dimension of the charging pile power response data to obtain the time-domain mapping matrix.

[0016] The generation module is used to perform particle swarm optimization calculations based on the time-domain mapping matrix and the fault risk assessment value to obtain a charging power allocation scheme, and generate a set of control instructions according to the charging power allocation scheme. The set of control instructions includes the power and time parameters of each charging pile.

[0017] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0018] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods described above.

[0019] In summary, the technical solution provided by this invention achieves efficient communication and state awareness between charging pile nodes by constructing an IoT topology and introducing a network connectivity matrix. It effectively extracts fault feature vectors using deep learning dimensionality reduction and node selection techniques, solving the problem of non-generalizability of fault location features in traditional methods and improving the ability to assess unknown fault locations. A hybrid neural network architecture including a spatial awareness subnetwork and a temporal analysis subnetwork is designed to achieve multi-dimensional analysis of the charging pile's operating state, significantly improving the accuracy of fault risk assessment. By constructing a temporal mapping matrix, a reduced-order mapping relationship is established between charging pile operating state variables, reducing computational complexity and improving energy optimization efficiency. A particle swarm optimization algorithm with a forgetting factor is introduced to achieve dynamic optimization of the charging power allocation scheme, effectively balancing charging efficiency and load balancing. Based on a two-stage robust programming model and an emergency uncertainty set, a complete fault emergency response mechanism is constructed, improving the operational stability of the charging pile system under fault conditions. By designing a hierarchical control command structure, seamless switching between normal operation mode and emergency mode is achieved, enhancing the fault adaptability and operational reliability of the charging pile system. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the steps of a smart operation and management method for charging piles based on the Internet of Things in one embodiment of the present invention;

[0021] Figure 2 This is a structural block diagram of an IoT-based smart operation and management system for charging piles in one embodiment of the present invention;

[0022] Figure 3 This is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0023] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0025] Reference Figure 1 This embodiment provides a smart operation and management method for charging piles based on the Internet of Things, including the following steps:

[0026] S1, construct the IoT topology for the charging pile nodes to obtain the node communication distance matrix and network connectivity matrix;

[0027] This process involves deploying IoT communication modules on charging piles. Each charging pile communication module is configured with communication parameters, including communication band, data transmission rate, and operating frequency. The device number and these communication parameters are recorded to form node deployment data, containing basic information about each communication module. For charging piles equipped with IoT communication modules, high-precision positioning technology is used to determine their spatial location. Using the Global Positioning System (GPS) or other high-precision spatial positioning methods, the geographic coordinates of each charging pile are acquired, including its longitude, latitude, and altitude. This data forms a geographic coordinate set, describing the distribution of all charging piles in the actual geographic space. The geographic coordinate set is then converted into standardized location data suitable for topology analysis. This conversion simplifies the complex geographic coordinate form into a three-dimensional Cartesian coordinate system directly applicable to mathematical analysis and modeling, reflecting the spatial distribution of charging piles. Based on the standardized location data, wireless communication links between nodes are constructed. By analyzing the spatial distance between each charging pile, a communication distance matrix is ​​generated, recording the distance information between each pair of nodes in the network. To more accurately describe communication capabilities, signal strength is calculated based on the propagation characteristics of wireless signals. By analyzing the distance between devices and the signal attenuation characteristics in space, the communication distance is converted into signal strength values, generating a signal strength matrix that reflects the actual communication quality between each pair of nodes. The signal strength matrix is ​​then normalized. The range of signal strength is limited to a standardized interval, making different strength values ​​more comparable, resulting in an initial connection strength matrix. A binary connection matrix is ​​generated by comparing the values ​​of the initial connection strength matrix with a preset connection threshold. If the connection strength of a pair of nodes exceeds the threshold, it is considered a valid connection, and the value in the matrix is ​​set to 1; otherwise, it is set to 0. A Laplace matrix transformation is then performed on the binary connection matrix. By comprehensively considering the number of connections to nodes and the specific connection relationships, a network connectivity matrix is ​​generated, reflecting the topology of the entire network, including the connectivity of each node and its importance in the overall network.

[0028] S2, based on the network connectivity matrix, performs deep learning dimensionality reduction and node selection to obtain the fault feature vector;

[0029] Specifically, the network connectivity matrix is ​​input into a deep autoencoder network for feature extraction. A deep autoencoder network is an unsupervised learning model that compresses and reconstructs data through an encoder and decoder structure. A deep autoencoder consists of three encoder layers and three decoder layers. The encoder layer-by-layer extracts deep features from the network connectivity matrix, mapping high-dimensional data to a low-dimensional latent space representation. The decoder reconstructs the input matrix from the low-dimensional representation, minimizing reconstruction error to ensure the representativeness of the extracted features. This step transforms the network connectivity matrix into an initial feature set containing key features. Principal component analysis (PCA) is performed on the initial feature set to reduce the dimensionality of the data, extracting principal component features to form a dimensionality-reduced feature matrix, effectively describing the main structural relationships between nodes in the network. A node importance evaluation function is constructed based on the dimensionality-reduced feature matrix. This function measures the importance of each node in the network by calculating its degree centrality and betweenness centrality. Degree centrality describes the direct connectivity of a node, while betweenness centrality reflects its role as an intermediary bridge in the global network. By combining these metrics, the importance of each node is quantified, resulting in a node score vector. A Top-K filtering operation is performed on the node score vectors to select the most important nodes, generating a target node index set. Top-K filtering sorts the node score vectors and selects the top K nodes with the highest scores as target nodes. These nodes occupy key positions in the network and have high priority and representativeness. Based on the target node index set, relevant subgraph structures are extracted from the network connectivity matrix to form a target node connection submatrix. The shortest path distance is calculated on the target node connection submatrix. The shortest path distance is an important indicator of the network distance between nodes, reflecting the proximity of target nodes in the topology. By calculating the shortest path distance on the connection submatrix, a node distance matrix is ​​generated. The node distance matrix is ​​normalized to transform all distance values ​​into a unified standard range, resulting in normalized distance features. The dimensionality-reduced feature matrix and the normalized distance features are then concatenated using a tensor operation to generate a fault feature vector.

[0030] S3, input the fault feature vector and charging pile operation status data into the hybrid neural network to calculate the risk and obtain the fault risk assessment value. The hybrid neural network includes a spatial perception subnetwork and a time series analysis subnetwork.

[0031] It should be noted that the fault feature vector is input into the spatial awareness subnetwork of the hybrid neural network. This subnetwork captures the spatial relationships between nodes in the network and structurally includes three graph attention layers, each containing a multi-head attention mechanism and residual connections. The multi-head attention mechanism can capture the interaction relationships between nodes from multiple perspectives, while the residual connections ensure that information is not excessively lost during feature transmission. In the first graph attention layer, the network performs preliminary processing on the input fault feature vector, extracting a feature representation with 128 dimensions; the second graph attention layer further compresses the features to 64 dimensions, extracting key patterns of spatial association between nodes through deeper learning; the third graph attention layer further compresses the features to 32 dimensions, obtaining the final spatial awareness output features. Simultaneously, the operating status data of the charging piles is processed, using a sliding window method to segment the data and generate a state sequence matrix. The sliding window segmentation process preserves the time-series characteristics of the operating status and makes the data more representative in time-series analysis. The generated state sequence matrix is ​​then input into the time-series analysis subnetwork of the hybrid neural network. This sub-network comprises two layers of Bidirectional Long Short-Term Memory (BiLSTM) networks. The first BiLSTM layer consists of 64 hidden units with an output dimension of 128. Its main function is to capture the forward and reverse dynamic characteristics of the time series and generate an initial temporal feature matrix. This initial temporal feature matrix is ​​then input into the second BiLSTM layer for feature extraction. The second layer contains 32 hidden units with an output dimension of 64. Through more refined temporal modeling, it generates temporal analysis output features, effectively summarizing the dynamic changes in charging pile operating status data. Feature fusion is performed on the spatial perception output features and the temporal analysis output features. These two features represent the spatial distribution characteristics and dynamic temporal characteristics of the charging pile network, respectively; their combination more comprehensively reflects the fault risk of the charging pile system. Through feature fusion, the two features are concatenated into a unified fused feature vector. This fused feature vector is then input into the network's multi-classification layer to calculate the fault risk probability. The multi-classification layer classifies the data according to different risk levels in the training data and outputs a fault risk probability distribution vector. The failure risk probability distribution vector contains probability values ​​corresponding to multiple risk levels, each value representing the likelihood of risk at that level. Multi-class classification normalizes the output to probabilistic form by introducing an activation function (such as softmax), ensuring that the sum of the probabilities of all risk levels is 1. A weighted risk score is calculated based on the failure risk probability distribution vector to obtain the final failure risk assessment value. The probability value of each risk level is multiplied by its corresponding risk weight coefficient, and then summed. The risk weight coefficient is set according to the severity of different risk levels in practical applications; for example, higher risk levels are assigned higher weights to focus more on the impact of severe failures. The weighted sum yields a comprehensive risk assessment value, quantifying the overall failure risk level of the current system.

[0032] S4. Normalize and reduce the dimension of the charging pile power response data to obtain the time-domain mapping matrix;

[0033] Specifically, power response data of charging piles is sampled. By collecting power data from different charging piles at specific time points, a power response matrix is ​​constructed. In this matrix, row vectors represent the identifiers of each charging pile, while column vectors correspond to the sampled time series points. Each element represents the power value of a charging pile at a specific time point. To capture the characteristics of charging pile power during dynamic changes, step response calculations are performed on the power response matrix. Step response is one of the core methods for analyzing system dynamics. By calculating the transient response to power changes, a step response dataset is generated. This dataset records the dynamic response process of charging piles when power changes, such as the transition behavior when power switches from a lower state to a higher state or vice versa. The step response dataset is normalized to map the numerical range of the data to a standard interval, such as [0,1], eliminating the influence of power amplitude differences between different charging piles and forming a standardized step response matrix. The standardized step response matrix is ​​then decomposed into a state space matrix and an output matrix. The former describes the internal state of the system, while the latter represents the external output of the system. State-space decomposition can divide a complex dynamic system into several independent subsystems, thus significantly reducing computational complexity while preserving key information in the system's dynamic characteristics. A dimensionality reduction transformation is performed on the state variable matrix, mapping the high-dimensional state space to a lower-dimensional subspace while preserving the main dynamic characteristics inherent in the original data to the greatest extent, resulting in a reduced-order state matrix. Simultaneously, combined with the output matrix, a reduced-order mapping function is constructed by designing a state observer to describe the dynamic relationship between the reduced-order state and the output of the system. The reduced-order mapping function is discretized on the time axis, converting continuous time dynamics into a series of discrete time nodes, generating a time-domain discrete sequence reflecting the change process of the reduced-order state at discrete time points. Matrix reconstruction operations are performed on the time-domain discrete sequence to construct the time-domain mapping matrix. In the time-domain mapping matrix, each element represents the state transition probability between adjacent time nodes, describing the state change trend of the charging pile between two time points. During the matrix reconstruction process, statistical characteristics between adjacent time nodes are extracted through analysis and calculation of the discrete sequence and converted into a quantified probabilistic form.

[0034] S5 performs particle swarm optimization calculations based on the time-domain mapping matrix and fault risk assessment values ​​to obtain a charging power allocation scheme. It then generates a set of control commands based on the charging power allocation scheme, which includes the power and time parameters of each charging pile.

[0035] Based on the time-domain mapping matrix and fault risk assessment values, a multi-objective optimization function is constructed. This function aims to simultaneously optimize the charging efficiency, load balancing, and system safety of charging piles. The charging efficiency sub-objective aims to increase the charging volume per unit time to meet users' demand for fast charging. The load balancing sub-objective minimizes the load differences among charging piles in the system, ensuring a more even power distribution and preventing individual charging piles from being overloaded or idle. The safety constraint sub-objective limits the power and time allocation of charging piles within a safe range, preventing equipment damage or safety accidents caused by power overload or abnormal charging duration. The multi-objective optimization function combines the dynamic characteristics of charging piles reflected in the time-domain mapping matrix with the potential node risks revealed in the fault risk assessment values. After constructing the multi-objective optimization function, weights are assigned to each sub-objective to form a comprehensive objective function. The weight allocation is based on the importance of different sub-objectives and the needs of actual application scenarios. For example, charging efficiency is prioritized during peak charging periods, while load balancing and safety need to be improved in scenarios where system stability is paramount. The comprehensive objective function unifies the different optimization objectives into a computable form by weighted summation of the sub-objectives. Based on the comprehensive objective function, a particle swarm optimizer with a forgetting factor is constructed, and the particle swarm P0 is initialized. The initial particle swarm contains M particles, each corresponding to a charging power allocation scheme, with a dimension of 2N, where N is the number of charging piles. The first N dimensions represent the power allocation value of each charging pile, and the last N dimensions represent the time allocation value of each charging pile. This structure allows the particles to simultaneously represent the power and time allocation schemes of the charging piles, providing flexibility for optimization calculations. After the initial particle swarm is generated, a fitness value set F is calculated for each particle, reflecting the merits of the scheme corresponding to the particle. By comparing the fitness values, the historical optimal solution ph and the global optimal solution pg of the particles are extracted. To avoid particles getting trapped in local optima and to better adapt to dynamic optimization needs, a forgetting factor λ is introduced to dynamically update the historical optimal solutions. The forgetting factor applies a certain decay weight to past optimal solutions, enabling particles to adjust their direction more quickly when exploring new solutions, improving the effect of global optimization. Velocity update calculations are performed on the initial particle swarm. The velocity update process combines historical optimal solutions, global optimal solutions, and the current particle position to ensure that particles simultaneously move towards both historical experience and the global best direction. The calculated particle velocity matrix is ​​superimposed with the current position to generate an updated particle position matrix. To ensure the effectiveness of the optimization results, boundary constraints are applied to the updated particle position matrix to ensure that the power allocation and time allocation values ​​for each charging station within the particle are within reasonable ranges and do not exceed the safe operating parameters of the charging stations. The global optimal solution is extracted from the updated particle position matrix to obtain the charging power allocation scheme, which includes the optimal power value and charging time period for each charging station.The global optimal solution is extracted based on a comparison of fitness values, selecting the particle that optimizes the overall objective function as the final solution. A set of control commands is generated based on the charging power allocation scheme. Each control command contains the power and time parameters for the corresponding charging pile, and these commands are sent to the charging pile's execution system to adjust the actual charging behavior.

[0036] Hierarchical clustering analysis was performed on historical fault data of charging piles. This data records various fault conditions that occurred during different operational phases of the system, including equipment malfunctions, power fluctuations, and communication interruptions. Hierarchical clustering analysis grouped this data according to fault characteristics, forming a high-dimensional fault feature matrix that reveals the potential correlations between different faults. Time-series probability prediction was then performed based on the fault feature matrix. By introducing time series analysis models (such as LSTM or ARIMA), the time dependence and trend characteristics in the historical data were captured, generating a 24-hour fault prediction sequence. This sequence indicates the possible fault distribution and probability of occurrence within the next day. The 24-hour fault prediction sequence and a set of control commands were input into a two-stage decision model generator to construct a two-stage robust programming model. The two-stage robust programming model divides system operation into two stages: the first stage is responsible for normal power allocation, ensuring efficient and balanced operation of charging piles under normal conditions; the second stage is used for emergency dispatch, designing dynamic adjustment schemes when faults occur. The objective function combines operating costs and emergency costs. Operating costs measure the system resource consumption under normal conditions, while emergency costs focus on the adjustment costs after a fault occurs. By jointly optimizing the variables in both stages, the model can reserve some adjustment space for emergency scenarios while meeting normal operation requirements, thus improving the system's robustness. To handle the uncertainty of emergency scenarios, the two-stage robust programming model is decomposed into emergency scenarios, generating an emergency uncertainty set U. A basic scenario set S0 is constructed, which contains all single-device failure scenarios, reflecting the basic situation when a single charging pile fails. A combined scenario set S1 is generated based on the correlation between devices, considering the possibility and impact of multiple devices failing simultaneously. By setting a probability threshold, the combined scenarios are filtered to remove scenarios with extremely low probability of occurrence, resulting in a more concise and practically meaningful emergency uncertainty set U. Based on the emergency uncertainty set U, a constraint generation subproblem is constructed, from which a new set of emergency constraints C is extracted. These constraints are used to limit the system's scheduling behavior when a failure occurs, such as the power adjustment range of the charging pile and the time switching window, to ensure the feasibility of the scheduling scheme in emergency scenarios. The constraints are applied to each instruction in the control instruction set for feasibility verification, resulting in a robust instruction subset R. The robust instruction subset contains control instructions that can maintain reliability under emergency scenarios, reflecting the system's robustness and flexibility in dealing with uncertain scenarios. Emergency dispatch strategies are designed based on a robust instruction subset R and an emergency uncertainty set U. The emergency dispatch strategy defines the system's automatic switching behavior when a fault is detected, such as dynamically adjusting the power allocation of the faulty device, reallocating charging tasks to surrounding charging stations, or shortening the execution time of certain tasks to reduce impact. These emergency dispatch strategies are encapsulated into a control instruction set to form the final target instruction set.Each instruction in the target instruction set is divided into a normal execution part and an emergency switching part. The normal execution part is used to ensure that the system runs as planned under fault-free conditions, while the emergency switching part is automatically triggered when a fault occurs, dynamically adjusting the operating state of the system.

[0037] In one example, an IoT topology is constructed for the charging pile nodes, yielding a node communication distance matrix and a network connectivity matrix, including:

[0038] The IoT communication module is deployed to each charging pile device to configure the communication parameters and obtain node deployment data. The node deployment data includes the device number and communication parameters of the IoT communication module.

[0039] Spatial positioning is performed on the charging pile equipment equipped with IoT communication modules to obtain a set of geographic coordinates, which includes the longitude, latitude and altitude values ​​of each charging pile.

[0040] The set of geographic coordinates is converted into a network topology coordinate system to obtain standardized location data. The standardized location data is characterized by three-dimensional rectangular coordinates (x, y, z) to represent the spatial distribution of nodes.

[0041] Wireless communication links are constructed for nodes in the standardized location data to obtain the node communication distance matrix, and the signal attenuation function is calculated based on the node communication distance matrix to obtain the signal strength matrix;

[0042] The signal strength matrix is ​​normalized by interval to obtain the initial connection strength matrix. The initial connection strength matrix is ​​then compared with the preset connection threshold to obtain the binarized connection matrix.

[0043] Perform a Laplacian matrix transformation on the binarized connection matrix to obtain the network connectivity matrix.

[0044] In this example, IoT communication modules are deployed on each charging pile device, and communication parameters are configured for each module. These parameters include communication frequency, transmission rate, signal strength, etc., to ensure that the modules can operate normally and communicate reliably with other nodes. After deployment, each module is assigned a unique device number (e.g., ...). Indicates the first The charging station's number was recorded, along with its communication parameters. Node deployment data is represented as This collection reflects the identification and communication capabilities of each charging station. Spatial positioning is performed on charging station devices equipped with IoT communication modules. Geographic coordinate data for each charging station, including longitude, is obtained through high-precision positioning technologies (such as GPS or differential positioning). The marker can accurately describe the physical location of the charging station in three-dimensional space, forming a complete set of geographic coordinates: in This represents the total number of charging stations. To facilitate network topology analysis, the geographic coordinate set is converted into standardized location data in a three-dimensional Cartesian coordinate system. The spherical coordinates (longitude and latitude) are converted to Cartesian coordinates using the following formula:

[0045] ;

[0046] ;

[0047] ;

[0048] in, It is the average radius of the Earth. It is the first The altitude of each node, and These are its longitude and latitude, respectively. The spatial location of each node is determined using these formulas. This indicates the generation of a standardized location dataset:

[0049] ;

[0050] After obtaining standardized location data, wireless communication links between nodes are constructed. The Euclidean distance between each pair of nodes is calculated using the following formula:

[0051] ;

[0052] in, Represents a node and nodes The spatial distance between nodes. A node communication distance matrix is ​​generated by calculating the distance between all node pairs. Each element Indicates the first The node and the first Distance between nodes. Based on the node communication distance matrix. Calculate the signal attenuation function and generate the signal strength matrix. Signal attenuation is expressed by the following formula:

[0053] ;

[0054] in, It is a node The emitted signal is at the node The intensity received at that location, It is a node The transmission power, It is the path loss index (generally ranging from 2 to 4, depending on the propagation environment). It is a node and The communication distance. Through this calculation, the signal strength matrix is ​​obtained. Each element This represents the communication signal strength between two nodes. To make the signal strength matrix data more intuitive and easier to process, it undergoes interval normalization. The normalization formula is:

[0055] ;

[0056] Normalized initial connection strength matrix All signal strength values ​​are compressed to the range [0, 1] to eliminate the influence of signal power differences between different nodes on the results. The initial connection strength matrix is ​​then used. With preset connection threshold A comparison operation is performed to determine whether a valid connection exists between the nodes. If Then the node is considered and nodes If a communication connection exists, it is considered to exist; otherwise, it is considered not to exist. This process generates a binary connection matrix. Its elements are defined as:

[0057] ;

[0058] For the binary connection matrix Perform a Laplacian matrix transformation to generate the network connectivity matrix. The Laplacian matrix is ​​defined as follows:

[0059] ;

[0060] in, It is a degree matrix, a diagonal matrix, whose diagonal elements Represents a node The degree, that is, the total number of connections to that node:

[0061] ;

[0062] The network connectivity matrix generated by Laplacian matrix transformation It can effectively reflect the connectivity characteristics of nodes in the network and their importance in the overall topology.

[0063] In one example, deep learning dimensionality reduction and node selection are performed based on the network connectivity matrix to obtain a fault feature vector, including:

[0064] The network connectivity matrix is ​​input into the deep autoencoder network for feature extraction to obtain the initial feature set. The deep autoencoder network contains a three-layer encoder and a three-layer decoder structure.

[0065] Principal component analysis is performed on the initial feature set to obtain a dimension-reduced feature matrix. A node importance evaluation function is then constructed based on the dimension-reduced feature matrix to obtain a node score vector. The node score vector is calculated using node degree centrality and betweenness centrality.

[0066] The node score vectors are subjected to Top-K filtering to obtain the target node index set. The subgraph structure is then extracted from the network connectivity matrix based on the target node index set to obtain the target node connection submatrix.

[0067] Calculate the shortest path distance for the target node connection submatrix to obtain the node distance matrix, and then normalize the node distance matrix to obtain the normalized distance feature;

[0068] Tensor concatenation operations are performed on the dimensionality-reduced feature matrix and the normalized distance feature to obtain the fault feature vector.

[0069] In this example, the network connectivity matrix The input is fed into a deep autoencoder network for feature extraction. A deep autoencoder network is an unsupervised learning model consisting of a three-layer encoder and a three-layer decoder. The encoder progressively maps high-dimensional data to a low-dimensional latent space to compress information and extract key features, while the decoder remaps the compressed features back to the original space to minimize reconstruction error. Assuming the input network connectivity matrix... ,in This represents the number of nodes and the size of the network. After passing through the first encoder layer, the data is mapped to a hidden layer, with an output dimension of... :

[0070] ;

[0071] in, It is a weight matrix. It is a bias vector. It is an activation function, such as ReLU. The second encoder will... This is then mapped to a lower-dimensional hidden layer, with dimension 1. :

[0072] ;

[0073] Similarly, after the third encoder layer, a low-dimensional feature representation of the latent space is obtained. The decoder part proceeds in reverse, converting... Gradually decode back to the original space, minimizing reconstruction error. To optimize network weights. Among them, This is the reconstructed connectivity matrix. The extracted latent features... Defined as the initial feature set, representing the key characteristics of the network. Principal component analysis (PCA) is performed on the initial feature set to reduce dimensionality and extract the principal features. PCA can identify the principal component directions of the data, which can explain the maximum variance in the data. Assume the initial feature set is a matrix. ,right Perform covariance matrix calculation:

[0074] ;

[0075] Among them, express The feature correlation is then calculated, followed by the eigenvalues ​​and eigenvectors of the covariance matrix, and the first eigenvalues ​​are selected. The largest eigenvalues ​​and their corresponding eigenvectors form a dimensionless feature matrix. The dimensionality-reduced feature matrix retains the main information of the network structure while significantly reducing data complexity. Based on the dimensionality-reduced feature matrix... Construct a node importance evaluation function to obtain the node score vector. The importance of nodes is measured using centrality metrics in graph theory, including degree centrality and betweenness centrality. Degree centrality... Represents a node The number of direct connections is defined as:

[0076] ;

[0077] in These are the elements of the binarized adjacency matrix, representing nodes. and Whether they are directly connected. Betweenness centrality. This reflects the role of a node in all shortest paths, and is defined as:

[0078] ;

[0079] in, It is a node and nodes The number of shortest paths between them These are the nodes passed through in these paths. The number of nodes. Through normalization, the two centrality indices are combined to obtain the comprehensive node score vector. Perform a Top-K selection on the node score vectors and select the node with the highest score. Each node is used as a target node to generate a target node index set. Based on the target node index set, from the network connectivity matrix... Extract the subgraph structure of the target node to obtain the target node connection submatrix. Connect the target node to the submatrix. Perform shortest path distance calculations to generate a distance matrix between nodes. Shortest path distance It is a node and The path weights and minimum path length between them. By analyzing... Normalization is performed to map the distance values ​​to the interval [0,1], generating a normalized distance feature matrix. Dimensionally reduced feature matrix and normalized distance feature matrix Tensor concatenation is performed. The concatenated tensors combine the structural features and spatial distance characteristics of the nodes to form the final fault feature vector. .

[0080] In one example, fault feature vectors and charging pile operating status data are input into a hybrid neural network for risk calculation to obtain a fault risk assessment value. The hybrid neural network includes a spatial perception subnetwork and a time series analysis subnetwork, including:

[0081] The fault feature vector is input into the spatial perception sub-network of the hybrid neural network for feature extraction. The spatial perception sub-network contains three graph attention layers. Each graph attention layer contains a multi-head attention mechanism and residual connections. The output dimension of the first graph attention layer is 128, the output dimension of the second graph attention layer is 64, and the output dimension of the third graph attention layer is 32, thus obtaining the spatial perception output features.

[0082] The charging pile operation status data is segmented by a sliding window to obtain a state sequence matrix. The state sequence matrix is ​​then input into the first bidirectional long short-term memory layer of the temporal analysis subnetwork in the hybrid neural network. The first bidirectional long short-term memory layer contains 64 hidden units and has an output dimension of 128, thus obtaining the initial temporal feature matrix.

[0083] The initial temporal feature matrix is ​​input into the second bidirectional long short-term memory layer of the temporal analysis subnetwork for feature extraction. The second bidirectional long short-term memory layer contains 32 hidden units and has an output dimension of 64, thus obtaining the temporal analysis output features.

[0084] Feature fusion is performed on the output features of spatial perception and the output features of temporal analysis to obtain a fused feature vector. Multi-classification calculation is then performed on the fused feature vector to obtain the fault risk probability distribution vector.

[0085] The weighted risk score is calculated based on the probability distribution vector of the failure risk to obtain the failure risk assessment value. The weighted risk score is calculated by summing the product of the probability value of each risk level and the corresponding risk weight coefficient.

[0086] In this example, the fault feature vector The input is fed into the spatial awareness subnetwork of the hybrid neural network. Indicates the number of nodes. This represents the feature dimension of each node. The spatially aware subnetwork extracts features progressively through three graph attention layers, each containing a multi-head attention mechanism and residual connections. The core of the graph attention layer lies in calculating the correlation between nodes and weighting and aggregating the features of neighboring nodes through the attention mechanism. For the , Graph attention of layers, nodes The output feature is defined as:

[0087] ;

[0088] in, Indicates the first The node at the th Features of the layer It is a node The neighborhood group, It is a weight matrix. It is a node and nodes The attention weights between individuals are calculated using the following formula:

[0089] ;

[0090] in, It is a learnable attention vector. This represents the feature concatenation operation. The output dimension of the first graph attention layer is 128, and the generated feature matrix is... Through similar calculations, the output dimensions of the second and third layers are 64 and 32, respectively, yielding the spatial perception output features. The charging pile operation status data is preprocessed, and a status sequence matrix is ​​generated by segmenting the data using a sliding window. ,in It is the number of time windows. This refers to the feature dimension within a single window (such as parameters like power and voltage). The sliding window method can capture the time-series characteristics of data while preserving the local features of each time segment. The state sequence matrix... The input is fed into the first bidirectional long short-term memory (BilSTM) layer of the temporal analysis subnetwork of the hybrid neural network. The first BiLSTM layer contains 64 hidden units and has an output dimension of 128. For each time step... The output of the bidirectional LSTM is calculated as follows:

[0091] ;

[0092] in, and These are the output features of the forward and backward LSTMs, respectively. The outputs of the entire sequence form the initial temporal feature matrix. The initial temporal feature matrix is ​​input into the second bidirectional long short-term memory layer for feature extraction. The second BiLSTM layer contains 32 hidden units and has an output dimension of 64, ultimately generating the temporal analysis output features. After completing spatial perception and temporal analysis, the two sets of features are fused. The spatial perception output features are then processed. and time series analysis output features By fusing features through feature concatenation, a fused feature vector is generated. The fused feature vectors are input into the multi-classification layer to calculate the failure risk probability. The multi-classification layer uses a softtmax activation function to transform the output into a probability distribution for each risk level.

[0093] ;

[0094] in, Risk level The probability, It is a classifier for levels The unnormalized score, This represents the total number of risk levels. The final generated failure risk probability distribution vector is: A weighted risk score is calculated based on the failure risk probability distribution vector to obtain the final failure risk assessment value. The weighted risk score is calculated using the following formula:

[0095] ;

[0096] in, Risk level The weight reflects the degree of impact of that level on the overall risk of the system. This represents the probability of the corresponding level. The evaluation value is calculated through weighted averages. Quantify the failure risk level of the charging pile system under the current condition.

[0097] In one example, the charging pile power response data is normalized and dimension-reduced to obtain a time-domain mapping matrix, including:

[0098] The power response data of the charging piles is sampled to obtain the power response matrix. The row vectors of the power response matrix represent the charging pile identifiers, and the column vectors represent the time series sampling points.

[0099] Step response calculation is performed on the power data in the power response matrix to obtain a step response dataset, which records the dynamic response process of the charging pile at the moment of power change.

[0100] Normalize the step response dataset to obtain the standardized step response matrix, and then perform state-space decomposition on the standardized step response matrix to obtain the state variable matrix and the output matrix.

[0101] The state variable matrix is ​​reduced in dimension to obtain a reduced state matrix. A state observer is designed based on the reduced state matrix and the output matrix to obtain a reduced mapping function.

[0102] The reduced-order mapping function is discretized on the time axis to obtain a time-domain discrete sequence. Then, a matrix reconstruction operation is performed on the time-domain discrete sequence to obtain a time-domain mapping matrix. Each element of the time-domain mapping matrix represents the state transition probability between adjacent time nodes.

[0103] In this example, the power response data of the charging pile is sampled to form a power response matrix. ,in Indicates the number of charging stations. This represents the number of time series sampling points. In the matrix... In this matrix, each row represents the power response record of a charging pile over the entire time series, while each column represents the power value of all charging piles at a specific time point. To analyze the dynamic response characteristics of charging piles during power changes, step response calculations are performed on the power response matrix. Step response is one of the core characteristics of system dynamics, used to describe the transient behavior of a system transitioning from one steady state to another. By calculating the response curves for power changes, a step response dataset is generated. Each set of data describes the step response process of the charging pile at a specific moment. For example, recording the system's response process. It satisfies the following form:

[0104] ;

[0105] in, and These represent the initial and final steady-state power values, respectively. It is the change in power. It is the time constant of the response speed. This refers to the time when the power change begins. The generated step response dataset is normalized to map the power response data of different charging piles to a standard range, such as [0,1], for unified processing. The normalized matrix... Represented as:

[0106] ;

[0107] in, It is the original step response data. and These are the minimum and maximum values ​​in the dataset, respectively. State-space decomposition is performed on the standardized step response matrix to obtain the state variable matrix. and output matrix The state-space model takes the following form:

[0108] ;

[0109] ;

[0110] in, It is a state variable. It is the state transition matrix. It is the input matrix. It is system input. It is an output variable. and These are the output matrix and the direct transfer matrix, respectively. The system is modeled using numerical methods, and the matrix is ​​extracted from the standardized matrix. Parameters, etc. To simplify the system, the state variable matrix... Dimensionality reduction is performed, and key features are extracted through singular value decomposition or principal component analysis to obtain the reduced-order state matrix. ,in After dimensionality reduction, a reduced-order mapping function is constructed by designing a state observer:

[0111] ;

[0112] ;

[0113] in, These are the state-space parameters after order reduction. The order-reduced mapping function is discretized on the time axis to generate a discrete time-domain sequence. The discretization process dynamically transforms continuous time into state values ​​at discrete time points. Let the discrete time step be... The discretized state equations are obtained as follows:

[0114] ;

[0115] in, It is a discrete state transition matrix. It is a discrete input matrix. Matrix reconstruction operations are performed on the discrete time-domain sequence to generate a time-domain mapping matrix. Elements of the time-domain mapping matrix Indicates at a time node and The state transition probability between states is calculated using the following formula:

[0116] ;

[0117] This matrix provides the dynamic behavior of charging piles at various time points, which can effectively support the analysis and optimization of power response.

[0118] In one example, particle swarm optimization is performed based on the time-domain mapping matrix and fault risk assessment values ​​to obtain a charging power allocation scheme. A set of control commands is then generated based on this scheme, containing power and time parameters for each charging station, including:

[0119] A multi-objective optimization function is constructed based on the time-domain mapping matrix and the fault risk assessment value. The multi-objective optimization function includes charging efficiency sub-objective, load balancing sub-objective and safety constraint sub-objective.

[0120] The weighted comprehensive objective function is obtained by calculating the weight allocation of the multi-objective optimization function.

[0121] A particle swarm optimizer with a forgetting factor is constructed based on a comprehensive objective function to obtain an initial particle swarm P0. The initial particle swarm P0 contains M particles, each particle has a dimension of 2N, where N is the number of charging piles, the first N dimensions represent power allocation, and the last N dimensions represent time allocation.

[0122] For each particle in the initial particle swarm P0, calculate the historical optimal solution ph and the global optimal solution pg to obtain the set of particle fitness values ​​F, and introduce a forgetting factor λ to dynamically update the historical optimal solution;

[0123] The initial particle swarm velocity is updated based on the forgetting factor λ to obtain the particle velocity matrix;

[0124] The particle velocity matrix is ​​superimposed with the current position to obtain the updated particle position matrix. Boundary constraints are then applied to the particle position matrix to ensure that the power and time parameters are within the effective range.

[0125] The global optimal solution is extracted from the updated particle position matrix to obtain the charging power allocation scheme, which includes the optimal charging power value and charging time period for each charging pile.

[0126] A set of control commands is generated based on the charging power allocation scheme. The set of control commands includes the power and time parameters of each charging pile.

[0127] In this example, based on the time-domain mapping matrix and failure risk assessment value A multi-objective optimization function is constructed. The core of multi-objective optimization is to simultaneously optimize multiple objective functions. The multi-objective optimization function includes three sub-objectives: charging efficiency, load balancing, and safety constraints. The charging efficiency sub-objective... The term used to maximize the total charging power of all charging stations is defined as:

[0128] ;

[0129] in, It is the first The power value of each charging station is shown in the negative sign, indicating that it is transformed into a minimization problem. Load balancing sub-objective To minimize the load imbalance of charging piles in the system, it is defined as:

[0130] ;

[0131] in, This is the average power of all charging stations. (Safety constraint sub-objective) To limit the allocation range of power and time, ensuring that parameters meet system safety requirements, it is defined in the form of a penalty function:

[0132] ;

[0133] in, and These are the maximum permissible values ​​for power and time, respectively. These three sub-objectives are combined into a single comprehensive objective function. To balance the importance of different objectives, weights are assigned to each sub-objective. The overall objective function is in the following form:

[0134] ;

[0135] in, By adjusting weights, certain objectives are prioritized, such as increasing the weight of charging efficiency during peak hours and focusing more on load balancing during off-peak hours. Based on the comprehensive objective function, a particle swarm optimization (PSO) system with a forgetting factor is constructed. The PSO optimizer gradually approaches the global optimum by simulating the movement of particles in the solution space. Initialization of a particle swarm is then performed. ,Include There are 3 particles, each with a dimension of 1. , among which the former The dimension indicates the power allocation of the charging station, followed by... The dimension represents the time allocation. Let the particle... The position is The speed is For each particle, calculate its fitness value. This refers to the value of the comprehensive objective function. Record the historical best solution for each particle. and the global optimal solution To enhance adaptability to dynamic environments, a forgetting factor is introduced. Dynamically update the historical best solution:

[0136] ;

[0137] The velocity and position of each particle are updated based on the forgetting factor. The velocity update formula is:

[0138] ;

[0139] in, It is inertial weight. It is a learning factor. It's a random number. The position update formula is:

[0140] ;

[0141] To ensure that the power and time allocation of the particles are within an effective range, boundary constraints are applied to the updated particle positions:

[0142] ;

[0143] ;

[0144] Extracting the global optimal solution through multiple iterations. This refers to the optimal power and time allocation scheme for all charging stations. Assume the final charging power allocation scheme is... The time allocation scheme is as follows Based on the charging power allocation scheme, a set of control commands is generated. Each command includes the charging station number, power parameters, and time parameters, for example:

[0145] ;

[0146] All the instructions together constitute the system's charging scheduling scheme, which is then executed by the charging piles.

[0147] In one example, the IoT-based smart operation and management method for charging piles also includes:

[0148] Hierarchical clustering analysis was performed on historical fault data of charging piles to obtain a fault feature matrix, and time series probability prediction was performed based on the fault feature matrix to obtain a 24-hour fault prediction sequence.

[0149] The 24-hour fault prediction sequence and control command set are input into the two-stage decision model generator to obtain the two-stage robust programming model. The first-stage variable of the two-stage robust programming model is the normal operation power allocation, and the second-stage variable is the emergency dispatch scheme. The objective function includes the operating cost item and the emergency cost item.

[0150] The two-stage robust programming model is decomposed into emergency scenarios to obtain the emergency uncertainty set U. The emergency uncertainty set U is generated through the following steps: First, a basic scenario set S0 is constructed, which contains all single device failure scenarios. Then, a combined scenario set S1 is generated based on the correlation between devices. Finally, the emergency uncertainty set U is obtained by filtering through probability thresholds.

[0151] Based on the emergency uncertainty set U, a constraint generation subproblem is constructed to obtain a new emergency constraint condition set C. The feasibility of each instruction in the control instruction set is verified by applying the emergency constraint condition set C to obtain a robust instruction subset.

[0152] An emergency scheduling strategy is designed based on a robust instruction subset R and an emergency uncertainty set U. The emergency scheduling strategy is then encapsulated into a control instruction set to obtain a target instruction set. Each instruction in the target instruction set contains a normal execution part and an emergency switching part. When a fault is detected, the corresponding emergency scheduling strategy is automatically triggered.

[0153] In this example, hierarchical clustering analysis is performed on the historical fault data of charging stations. It is assumed that the historical fault data includes... A charging station in the past Fault record matrix at each time point Each element Indicates the first Each charging station in time The fault intensity (e.g., number of faults or severity) is used. Based on this data, a hierarchical clustering algorithm is employed to group charging stations with similar fault characteristics into one category by calculating the similarity between them (e.g., Euclidean distance or cosine similarity). The clustering similarity metric is defined as:

[0154] Similarity ;

[0155] By constructing a clustering tree and setting an appropriate cutting threshold, we obtain... There are three clusters, each representing a group of charging piles with similar fault characteristics. Based on the clustering results, a fault feature matrix is ​​constructed. Each row represents the change in the average fault intensity of a cluster over time. Based on the fault feature matrix... Perform time-series probabilistic prediction to forecast failure scenarios over the next 24 hours. Employ a time-series prediction model (such as LSTM or ARIMA), with an input matrix... The trained model captures the dynamic characteristics of time series data and outputs a 24-hour fault prediction sequence. Each element Indicates the first The clustering in the th case is at the th case. Hourly failure probability. Combining the 24-hour failure prediction sequence with the control command set. The input is fed into a two-stage decision model generator to construct a two-stage robust programming model. The first-stage variable of the two-stage robust programming model is the normal operating power allocation, denoted as... ,in It is the first The power allocation for each charging station. The second-stage variable is the emergency dispatch plan, denoted as... ,in It is the first The power adjustment amount of each charging pile in an emergency scenario. The objective function of the model consists of operating cost and emergency cost terms:

[0156] ;

[0157] in, It is the unit cost of operating power. This represents the unit cost of emergency adjustments. The goal is to simultaneously optimize the costs of normal operation and emergency response. To improve robustness, the two-stage robust programming model is decomposed into emergency scenarios to generate an emergency uncertainty set. Build a basic scene set It includes all single-device failure scenarios, each scenario Simulates the fault behavior of only one charging station. Generates a set of combined scenarios based on the correlation between devices. Correlation is calculated using historical fault data, such as charging pile data. and The correlation is expressed as:

[0158] Correlation ;

[0159] Combined failure scenarios are generated by selecting highly correlated device combinations. A probability threshold is then set. The probability of screening occurring is higher than In the scenario, an emergency uncertainty set is obtained. Based on emergency uncertainty set Construct a constraint generation subproblem to generate a new set of emergency constraints. The constraints include boundary constraints for power and time allocation:

[0160] ;

[0161] Simultaneously, considering the changes in power demand in uncertain scenarios, specific power constraints are generated for each scenario:

[0162] ;

[0163] in, Representing a scene Next Minimum requirements for a single charging station. Apply emergency constraint conditions to each instruction in the control instruction set. Feasibility verification is performed, and instructions that cannot meet the constraints are removed to obtain a robust subset of instructions. Based on a robust instruction subset and emergency uncertainty set An emergency dispatch strategy is designed. This strategy defines power adjustment schemes based on different scenarios to ensure the system can respond quickly in the event of a fault. For example, when a charging station is detected... In case of failure, reduce its power distribution. At the same time, dynamically increase the power of surrounding charging piles. To meet overall requirements, these emergency strategies are encapsulated into a set of control instructions, generating a target instruction set. Each instruction in the target instruction set includes a normal execution part and an emergency switching part, in the following form:

[0164] normal: Emergency: ;

[0165] When a fault is detected, an emergency dispatch strategy is automatically triggered to adjust power and time.

[0166] Reference Figure 2 This embodiment provides an IoT-based intelligent operation and management system for charging piles, including:

[0167] Module 1 is used to construct the IoT topology of the charging pile nodes and obtain the node communication distance matrix and network connectivity matrix.

[0168] Filtering module 2 is used for deep learning dimensionality reduction and node filtering based on the network connectivity matrix to obtain fault feature vectors;

[0169] Calculation module 3 is used to input fault feature vectors and charging pile operation status data into a hybrid neural network for risk calculation to obtain fault risk assessment values. The hybrid neural network includes a spatial perception subnetwork and a time series analysis subnetwork.

[0170] Mapping module 4 is used to normalize and reduce the dimension of the charging pile power response data to obtain a time-domain mapping matrix;

[0171] The generation module 5 is used to perform particle swarm optimization calculations based on the time-domain mapping matrix and fault risk assessment values ​​to obtain a charging power allocation scheme, and generate a set of control instructions based on the charging power allocation scheme. The set of control instructions includes the power and time parameters of each charging pile.

[0172] In this embodiment, the specific implementation of each unit in the above system embodiment is described in the above method embodiment, and will not be repeated here.

[0173] Reference Figure 3 This invention also provides a computer device, which can be a server, and its internal structure can be as follows: Figure 3 As shown, the computer device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores the data corresponding to this embodiment. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0174] Those skilled in the art will understand that Figure 3 The structures shown are merely block diagrams of some structures related to the present invention and do not constitute a limitation on the computer devices on which the present invention is applied.

[0175] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0176] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the present invention and embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0177] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0178] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A smart operation and management method for charging piles based on the Internet of Things, characterized in that, Includes the following steps: IoT topology construction is performed on charging pile nodes to obtain node communication distance matrix and network connectivity matrix; Specifically, the process includes: deploying IoT communication modules to each charging pile device and configuring communication parameters to obtain node deployment data, which includes the device number and communication parameters of the IoT communication module; performing spatial positioning on the charging pile device equipped with the IoT communication module to obtain a set of geographic coordinates, which includes the longitude, latitude, and altitude values ​​of each charging pile; converting the set of geographic coordinates into a network topology coordinate system to obtain standardized location data, which represents the spatial distribution of nodes using three-dimensional rectangular coordinates (x, y, z); constructing wireless communication links for the nodes in the standardized location data to obtain a node communication distance matrix, and calculating a signal attenuation function based on the node communication distance matrix to obtain a signal strength matrix; performing interval normalization on the signal strength matrix to obtain an initial connection strength matrix, and comparing the initial connection strength matrix with a preset connection threshold to obtain a binary connection matrix; and performing a Laplace matrix transformation on the binary connection matrix to obtain a network connectivity matrix. Based on the network connectivity matrix, deep learning dimensionality reduction and node selection are performed to obtain a fault feature vector. Specifically, this includes: inputting the network connectivity matrix into a deep autoencoder network for feature extraction, obtaining an initial feature set; the deep autoencoder network comprising a three-layer encoder and a three-layer decoder structure; performing principal component analysis on the initial feature set to obtain a dimensionality-reduced feature matrix; constructing a node importance evaluation function based on the dimensionality-reduced feature matrix to obtain a node score vector, which is calculated using node degree centrality and betweenness centrality; performing Top-K filtering on the node score vector to obtain a target node index set; extracting a subgraph structure from the network connectivity matrix based on the target node index set to obtain a target node connection submatrix; calculating the shortest path distance on the target node connection submatrix to obtain a node distance matrix; normalizing the node distance matrix to obtain a normalized distance feature; and performing tensor concatenation on the dimensionality-reduced feature matrix and the normalized distance feature to obtain the fault feature vector. The fault feature vector and the charging pile operating status data are input into a hybrid neural network for risk calculation to obtain a fault risk assessment value. The hybrid neural network includes a spatial perception subnetwork and a time series analysis subnetwork. The power response data of the charging pile is normalized and dimensionally reduced to obtain the time-domain mapping matrix; Based on the time-domain mapping matrix and the fault risk assessment value, particle swarm optimization calculation is performed to obtain a charging power allocation scheme. A set of control commands is then generated according to the charging power allocation scheme, and the set of control commands includes the power and time parameters of each charging pile.

2. The IoT-based intelligent operation and management method for charging piles according to claim 1, characterized in that, The fault feature vector and charging pile operating status data are input into a hybrid neural network for risk calculation to obtain a fault risk assessment value. The hybrid neural network includes a spatial perception subnetwork and a time series analysis subnetwork, including: The fault feature vector is input into the spatial perception sub-network of the hybrid neural network for feature extraction. The spatial perception sub-network contains three graph attention layers. Each graph attention layer contains a multi-head attention mechanism and residual connections. The output dimension of the first graph attention layer is 128, the output dimension of the second graph attention layer is 64, and the output dimension of the third graph attention layer is 32, thus obtaining the spatial perception output features. The charging pile operation status data is segmented by a sliding window to obtain a state sequence matrix. The state sequence matrix is ​​then input into the first bidirectional long short-term memory layer of the time series analysis subnetwork in the hybrid neural network. The first bidirectional long short-term memory layer contains 64 hidden units and has an output dimension of 128 to obtain the initial time series feature matrix. The initial temporal feature matrix is ​​input into the second bidirectional long short-term memory layer of the temporal analysis sub-network for feature extraction. The second bidirectional long short-term memory layer contains 32 hidden units and has an output dimension of 64, thus obtaining the temporal analysis output features. The spatial perception output features and the temporal analysis output features are fused to obtain a fused feature vector, and the fused feature vector is then subjected to multi-classification calculation to obtain a fault risk probability distribution vector. The weighted risk score is calculated based on the fault risk probability distribution vector to obtain the fault risk assessment value. The weighted risk score is calculated by summing the product of the probability value of each risk level and the corresponding risk weight coefficient.

3. The intelligent operation and management method for charging piles based on the Internet of Things according to claim 2, characterized in that, The normalization and dimensionality reduction mapping of the charging pile power response data to obtain the time-domain mapping matrix includes: The power response data of the charging piles is sampled to obtain a power response matrix. The row vectors of the power response matrix represent the charging pile identifiers, and the column vectors represent the time series sampling points. A step response calculation is performed on the power data in the power response matrix to obtain a step response dataset, which records the dynamic response process of the charging pile at the instant of power change. The step response dataset is normalized to obtain a standardized step response matrix, and the standardized step response matrix is ​​decomposed into a state space to obtain a state variable matrix and an output matrix. The state variable matrix is ​​subjected to a dimensionality reduction transformation to obtain a reduced-order state matrix. A state observer is then designed based on the reduced-order state matrix and the output matrix to obtain a reduced-order mapping function. The reduced-order mapping function is discretized on the time axis to obtain a time-domain discrete sequence. Then, a matrix reconstruction operation is performed on the time-domain discrete sequence to obtain a time-domain mapping matrix. Each element of the time-domain mapping matrix represents the state transition probability between adjacent time nodes.

4. The intelligent operation and management method for charging piles based on the Internet of Things according to claim 3, characterized in that, The process involves particle swarm optimization based on the time-domain mapping matrix and the fault risk assessment value to obtain a charging power allocation scheme. A set of control commands is then generated based on this scheme. This set of control commands includes power and time parameters for each charging station, including: A multi-objective optimization function is constructed based on the time-domain mapping matrix and the fault risk assessment value. The multi-objective optimization function includes a charging efficiency sub-objective, a load balancing sub-objective, and a safety constraint sub-objective. The weighted comprehensive objective function is obtained by calculating the weight allocation of the multi-objective optimization function. Based on the comprehensive objective function, a particle swarm optimizer with a forgetting factor is constructed to obtain an initial particle swarm P0. The initial particle swarm P0 contains M particles, each particle having a dimension of 2N, where N is the number of charging piles, the first N dimensions represent power allocation, and the last N dimensions represent time allocation. For each particle in the initial particle swarm P0, calculate the historical optimal solution ph and the global optimal solution pg to obtain the particle fitness value set F, and introduce a forgetting factor λ to dynamically update the historical optimal solution; Based on the forgetting factor λ, the initial particle swarm is updated to obtain the particle velocity matrix. The particle velocity matrix is ​​superimposed with the current position to obtain the updated particle position matrix, and the particle position matrix is ​​subjected to boundary constraint processing to ensure that the power and time parameters are within the effective range. The global optimal solution is extracted from the updated particle position matrix to obtain a charging power allocation scheme, which includes the optimal charging power value and charging time period for each charging pile. A set of control commands is generated based on the charging power allocation scheme. The set of control commands includes the power and time parameters of each charging pile.

5. The intelligent operation and management method for charging piles based on the Internet of Things according to claim 4, characterized in that, The IoT-based intelligent operation and management method for charging piles also includes: Hierarchical clustering analysis is performed on historical fault data of charging piles to obtain a fault feature matrix, and time-series probability prediction is performed based on the fault feature matrix to obtain a 24-hour fault prediction sequence. The 24-hour fault prediction sequence and the control command set are input into the two-stage decision model generator to obtain a two-stage robust programming model. The first-stage variable of the two-stage robust programming model is the normal operation power allocation, and the second-stage variable is the emergency dispatch scheme. The objective function includes the operating cost item and the emergency cost item. The two-stage robust programming model is decomposed into emergency scenarios to obtain an emergency uncertainty set U. The emergency uncertainty set U is generated through the following steps: first, a basic scenario set S0 is constructed, which contains all single device failure scenarios; then, a combined scenario set S1 is generated based on the correlation between devices; and finally, the emergency uncertainty set U is obtained by filtering through probability thresholds. Based on the emergency uncertainty set U, a constraint generation subproblem is constructed to obtain a new emergency constraint condition set C. The feasibility of each instruction in the control instruction set is verified by applying the emergency constraint condition set C to obtain a robust instruction subset. An emergency dispatch strategy is designed based on the robust instruction subset R and the emergency uncertainty set U, and the emergency dispatch strategy is encapsulated into the control instruction set to obtain the target instruction set. Each instruction in the target instruction set contains a normal execution part and an emergency switching part. When a fault is detected, the corresponding emergency dispatch strategy is automatically triggered.

6. A smart operation and management system for charging piles based on the Internet of Things, characterized in that, The system comprises: steps for implementing the method according to any one of claims 1 to 5; The module is used to construct the IoT topology of charging pile nodes, and obtain the node communication distance matrix and network connectivity matrix. The filtering module is used to perform deep learning dimensionality reduction and node filtering based on the network connectivity matrix to obtain a fault feature vector. The calculation module is used to input the fault feature vector and the charging pile operation status data into the hybrid neural network for risk calculation to obtain the fault risk assessment value. The hybrid neural network includes a spatial perception subnetwork and a time series analysis subnetwork. The mapping module is used to normalize and reduce the dimension of the charging pile power response data to obtain the time-domain mapping matrix. The generation module is used to perform particle swarm optimization calculations based on the time-domain mapping matrix and the fault risk assessment value to obtain a charging power allocation scheme, and generate a set of control instructions according to the charging power allocation scheme. The set of control instructions includes the power and time parameters of each charging pile.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images