Power equipment control method and system based on internet of things

Abstract

This invention belongs to the field of equipment control technology, specifically relating to a power equipment control method and system based on the Internet of Things (IoT), comprising the following steps: S1, acquiring real-time operating data of nodes of each power device in the IoT, constructing a second-order state tensor representing the operating state of each node; projecting the second-order state tensor onto a low-dimensional subspace defined by basis vectors to obtain a low-dimensional state descriptor, wherein the basis vectors are adaptively updated based on the historical state covariance matrix of all nodes; calculating the Wasserstein distance of the low-dimensional state descriptor sequences between nodes, and generating a weighted transition matrix for parameter aggregation between nodes by combining the communication link quality; S2, if the condition number of the local state covariance matrix of a node exceeds a first threshold, then local optimization is initiated. This invention improves the convergence speed and accuracy of the optimization algorithm, and achieves more stable collaborative control of power equipment groups through a two-level control mechanism combining local optimization and global correction.

Description

Technical Field

[0001] This invention belongs to the field of equipment control technology, specifically relating to a power equipment control method and system based on the Internet of Things. Background Technology

[0002] With the integration of IoT technology into modern power systems, power equipment nodes are interconnected, enabling the perception and control of the power grid's operating status. However, the operating data generated by each equipment node exhibits characteristics of high dimensionality, heterogeneity, and strong dynamism. Control methods based on simplified models struggle to express the complex operating states and nonlinear relationships of the equipment. Furthermore, the complex IoT communication environment, with uncertainties such as latency and packet loss, places higher demands on the real-time performance of control strategies.

[0003] Existing power equipment control methods mostly employ centralized architectures or traditional distributed collaborative strategies. Centralized methods suffer from high communication overhead, computational bottlenecks, and single-point-of-failure risks, making them difficult to adapt to the expansion needs of large-scale equipment networks. While distributed methods improve system flexibility, they often lead to decreased control accuracy when processing high-dimensional data due to insufficient information representation. Furthermore, they often fail to adequately consider the dynamic differences in node states and the actual quality of communication links, affecting the efficiency of parameter coordination and control effectiveness. Moreover, simply vectorizing equipment operating data results in the loss of the equipment's inherent structural information. Regarding inter-node coordination, effectively measuring the differences between different node state sequences and adaptively aggregating parameters based on these differences is crucial for improving distributed control performance. Using conventional gradient optimization algorithms in Euclidean space not only violates the inherent manifold constraints of the data but may also lead to slow convergence or even non-convergence. Therefore, a novel power equipment control method capable of adapting to dynamic changes and performing efficient optimization is urgently needed to achieve the safe and stable operation of power systems in the Internet of Things (IoT) environment. Summary of the Invention

[0004] This invention provides a power equipment control method and system based on the Internet of Things, in order to solve the technical problems of the defects of the centralized architecture in the power equipment control of the prior art and the shortcomings of the traditional distributed method in terms of high-dimensional data processing, dynamic difference adaptation, manifold constraint preservation and optimization efficiency.

[0005] In a first aspect, the present invention provides a power equipment control method based on the Internet of Things, comprising the following steps:

[0006] S1. Obtain real-time operating data of each power device node in the Internet of Things, and construct a second-order state tensor representing the operating state of each node; project the second-order state tensor onto a low-dimensional subspace defined by basis vectors to obtain a low-dimensional state descriptor, and adaptively update the basis vectors according to the historical state covariance matrix of all nodes; calculate the Wasserstein distance between the low-dimensional state descriptor sequences of nodes, and generate a weighted transition matrix for parameter aggregation between nodes by combining the communication link quality.

[0007] S2, if the condition number of the local state covariance matrix of a node exceeds the first threshold, local optimization is initiated: the control parameters of neighboring nodes are aggregated using the weighted transition matrix, and the node's control parameters are updated through exponential mapping in combination with its own Riemann gradient; if the mean Wasserstein distance between global nodes exceeds the second threshold, global correction is initiated: the sparse optimization model is solved by the node with the highest betweenness centrality in the network to obtain global correction parameters, which are then distributed to each node to adjust the node's control parameters.

[0008] S3 decodes the updated control parameters into specific control instructions and executes them.

[0009] Furthermore, real-time operational data of each power device node in the Internet of Things is acquired, and a second-order state tensor representing the node's operational state is constructed for each node, including:

[0010] The voltage U, current I, frequency f, active power P, and reactive power Q of each power equipment node are collected as a real-time operating data sequence over 50 consecutive sampling time points.

[0011] The data sequence of the five physical quantities collected is used as the five-dimensional state vector of the node;

[0012] Calculate the covariance matrix of the five-dimensional state vector of the node at 50 sampling time points, and use the covariance matrix as a second-order state tensor.

[0013] Furthermore, the basis vectors are adaptively updated based on the historical state covariance matrix of all nodes, including:

[0014] Collect the historical state covariance matrices of all nodes over the past 100 periods, calculate the arithmetic mean of the historical state covariance matrices, and obtain the global average covariance matrix.

[0015] Perform eigenvalue decomposition on the global average covariance matrix;

[0016] Select the eigenvectors corresponding to the top 10 largest eigenvalues ​​as the updated basis vectors.

[0017] Furthermore, the Wasserstein distance between the low-dimensional state description sequences of the nodes is calculated, and combined with the communication link quality, a weighted transition matrix for parameter aggregation between nodes is generated, including:

[0018] Get the packet loss rate of bidirectional communication between node i and its neighbor node j. and average round-trip delay Calculate the communication link quality factor. ;

[0019] Obtain the Wasserstein distance between the low-dimensional state description subsequences of node i and its neighbor node j. ;

[0020] According to the formula Calculate the original weights of neighbor node j with respect to node i. ;

[0021] Normalize the original weights of all neighboring nodes of node i to obtain the elements in the i-th row of the weighted transition matrix.

[0022] Furthermore, if the condition number of a node's local state covariance matrix exceeds a first threshold, local optimization is initiated: the control parameters of neighboring nodes are aggregated using a weighted transition matrix, and the node's control parameters are updated via an exponential mapping, combining the node's own Riemann gradient, including:

[0023] When the condition number of the local state covariance matrix of node i exceeds the first threshold, the node performs the following operation:

[0024] The aggregation parameters are obtained by aggregating the control parameters of all neighboring nodes of a node using a weighted transition matrix. ;

[0025] Calculate its own control parameters Riemann gradient ;

[0026] Formulas combining exponential and logarithmic mappings:

[0027] Update the control parameters, where α is the learning rate, and... For exponential mapping, For logarithmic mapping, These are the updated control parameters.

[0028] Furthermore, if the mean Wasserstein distance between global nodes exceeds the second threshold, a global correction is initiated: the node with the highest betweenness centrality in the network solves the sparse optimization model to obtain global correction parameters, which are then distributed to each node to adjust the node's control parameters, including:

[0029] When the arithmetic mean of the Wasserstein distances between global nodes exceeds the second threshold, perform the following operation:

[0030] Calculate the betweenness centrality of all nodes in the network topology, and select the node with the highest betweenness centrality as the center node;

[0031] The central node constructs a LASSO regression model: , where A is the global state feature matrix, b is the ideal state vector, x is the global correction parameter to be solved, and λ is the regularization coefficient;

[0032] Solve the LASSO regression model to obtain sparse global correction parameters x, and then distribute the global correction parameters.

[0033] Secondly, the power equipment control system based on the Internet of Things includes the following modules:

[0034] The projection and aggregation module is used to acquire real-time operating data of nodes of various power devices in the Internet of Things, and construct a second-order state tensor representing the operating state of each node; project the second-order state tensor onto a low-dimensional subspace defined by basis vectors to obtain a low-dimensional state descriptor, and the basis vectors are adaptively updated according to the historical state covariance matrix of all nodes; calculate the Wasserstein distance of the low-dimensional state descriptor sequences between nodes, and generate a weighted transition matrix for parameter aggregation between nodes by combining the communication link quality;

[0035] The correction module is used to initiate local optimization if the condition number of the local state covariance matrix of a node exceeds the first threshold: it aggregates the control parameters of neighboring nodes using a weighted transition matrix, combines its own Riemann gradient, and updates the node's control parameters through exponential mapping; if the mean Wasserstein distance between global nodes exceeds the second threshold, it initiates global correction: the node with the highest betweenness centrality in the network solves the sparse optimization model to obtain global correction parameters, which are then distributed to each node to adjust the node's control parameters.

[0036] The distribution module is used to decode the updated control parameters into specific control commands and execute them.

[0037] Furthermore, real-time operational data of each power device node in the Internet of Things is acquired, and a second-order state tensor representing the node's operational state is constructed for each node, including:

[0038] The voltage U, current I, frequency f, active power P, and reactive power Q of each power equipment node are collected as a real-time operating data sequence over 50 consecutive sampling time points.

[0039] The data sequence of the five physical quantities collected is used as the five-dimensional state vector of the node;

[0040] Calculate the covariance matrix of the five-dimensional state vector of the node at 50 sampling time points, and use the covariance matrix as a second-order state tensor.

[0041] Furthermore, the basis vectors are adaptively updated based on the historical state covariance matrix of all nodes, including:

[0042] Collect the historical state covariance matrices of all nodes over the past 100 periods, calculate the arithmetic mean of the historical state covariance matrices, and obtain the global average covariance matrix.

[0043] Perform eigenvalue decomposition on the global average covariance matrix;

[0044] Select the eigenvectors corresponding to the top 10 largest eigenvalues ​​as the updated basis vectors.

[0045] Furthermore, the Wasserstein distance between the low-dimensional state description sequences of the nodes is calculated, and combined with the communication link quality, a weighted transition matrix for parameter aggregation between nodes is generated, including:

[0046] Get the packet loss rate of bidirectional communication between node i and its neighbor node j. and average round-trip delay Calculate the communication link quality factor. ;

[0047] Obtain the Wasserstein distance between the low-dimensional state description subsequences of node i and its neighbor node j. ;

[0048] According to the formula Calculate the original weights of neighbor node j with respect to node i. ;

[0049] Normalize the original weights of all neighboring nodes of node i to obtain the elements in the i-th row of the weighted transition matrix.

[0050] The beneficial effects are as follows: This invention preserves the inherent structural information and statistical characteristics of the data by constructing second-order state tensors for each node of the power equipment, reducing information loss caused by vectorization methods. By projecting the second-order state tensors onto a continuously updated low-dimensional subspace, the computational complexity of data processing and the burden on network communication are reduced. At the node collaboration level, the Wasserstein distance is used to measure the differences in node state sequences, and the aggregation weight is determined in conjunction with the actual quality of the communication link. This ensures that parameter interactions between nodes consider not only the similarity of states but also the reliability of information transmission, improving the efficiency of distributed collaboration. Furthermore, the update of control parameters is performed on a Riemannian manifold, preserving the inherent non-Euclidean geometry of the state data, improving the convergence speed and accuracy of the optimization algorithm. Through a two-level control mechanism combining local optimization and global correction, more stable collaborative control of the power equipment group is achieved. Attached Figure Description

[0051] Figure 1 This is a flowchart of an IoT-based power equipment control method. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] An embodiment of the Internet of Things-based power equipment control method provided by this invention:

[0054] like Figure 1 As shown, the IoT-based power equipment control method includes the following steps:

[0055] S1. Obtain real-time operating data of each power device node in the Internet of Things (IoT) and construct a second-order state tensor representing the operating state of each node. Project the second-order state tensor onto a low-dimensional subspace defined by basis vectors to obtain a low-dimensional state descriptor. The basis vectors are adaptively updated based on the historical state covariance matrix of all nodes. Calculate the Wasserstein distance between the low-dimensional state descriptor sequences of nodes and, combined with the communication link quality, generate a weighted transition matrix for parameter aggregation between nodes.

[0056] Multidimensional operating indicators of each power equipment node are continuously collected within a continuous time window, such as node voltage amplitude, current amplitude, active power, reactive power, and frequency, forming a multidimensional time series data matrix. Based on this multidimensional time series data matrix, a sample covariance matrix is ​​calculated. The sample covariance matrix is ​​a second-order state tensor representing the recent dynamic operating characteristics of the nodes. The diagonal elements of the tensor represent the fluctuation degree of each indicator, while the off-diagonal elements represent the correlation between different indicators.

[0057] We collect the historical second-order state tensors generated by all power device nodes in the Internet of Things (IoT) over a relatively long period of time, calculate the arithmetic mean of these second-order state tensors, and obtain a global average state covariance matrix. We then perform eigenvalue decomposition on the global average state covariance matrix, selecting the eigenvectors corresponding to the k largest eigenvalues. These mutually orthogonal eigenvectors form a set of basis vectors. We then left-multiply the current second-order state tensor of each node by the transpose of this set of basis vectors, and then right-multiply by the same set of basis vectors to perform a projection operation, obtaining a k×k low-dimensional state descriptor. This process is equivalent to performing principal component analysis on a Stiefel manifold.

[0058] The low-dimensional state descriptors obtained by a node in the most recent few time steps are treated as a sequence, and each low-dimensional state descriptor is considered as a covariance matrix of a zero-mean Gaussian distribution. The difference between the state sequences of any two nodes is quantified by calculating the second-order Wasserstein distance between their latest state descriptors. Simultaneously, the communication link quality is evaluated by measuring the round-trip time and packet loss rate of network packets between nodes, mapping the data to a quality factor ranging from 0 to 1. The weights from node i to node j are determined by the product of the communication quality factor and the negative exponential function of the Wasserstein distance. All weights constitute a weighted transition matrix, which is then normalized.

[0059] In an optional embodiment, real-time operating data of nodes of various power devices in the Internet of Things are obtained, and a second-order state tensor representing the operating state of each node is constructed, including:

[0060] The voltage U, current I, frequency f, active power P, and reactive power Q of each power equipment node are collected as a real-time operating data sequence over 50 consecutive sampling time points.

[0061] The data sequence of the five physical quantities collected is used as the five-dimensional state vector of the node;

[0062] Calculate the covariance matrix of the five-dimensional state vector of the node at 50 sampling time points, and use the covariance matrix as a second-order state tensor.

[0063] Taking a transformer node in the power grid as an example, 50 data points were continuously collected within a 500ms time window. Each data point contained five key physical quantities, such as voltage of 110.2kV, current of 205A, frequency of 50.01Hz, active power of 30MW, and reactive power of 4MW. For the transformer node, five real-time operational data sequences, each containing 50 data points, were obtained.

[0064] At each sampling time point, the measured values ​​of five physical quantities collectively constitute a five-dimensional state vector. For example, in the first 10 ms, the vector is 110.2, 205, 50.01, 30, 4. Within 500 ms, the transformer node generates 50 such five-dimensional state vectors. A 5×5 covariance matrix is ​​calculated based on these 50 five-dimensional state vectors. The diagonal elements of the covariance matrix represent the variance of individual physical quantities such as voltage and current over the observation time, while the off-diagonal elements represent the correlation between different physical quantities, such as the linear relationship between voltage and reactive power. The covariance matrix is ​​used as a second-order state tensor representing the operating characteristics of the transformer node.

[0065] In an optional embodiment, the basis vectors are adaptively updated based on the historical state covariance matrix of all nodes, including:

[0066] Collect the historical state covariance matrices of all nodes over the past 100 periods, calculate the arithmetic mean of the historical state covariance matrices, and obtain the global average covariance matrix.

[0067] Perform eigenvalue decomposition on the global average covariance matrix;

[0068] Select the eigenvectors corresponding to the top 10 largest eigenvalues ​​as the updated basis vectors.

[0069] Suppose a power grid monitoring system with 1000 nodes, where each node generates a second-order state tensor, i.e., a covariance matrix, per second. To achieve adaptive model updates, all historical state covariance matrices generated by all 100 nodes in the network over the past 100 seconds are collected, totaling 10,000 matrices. An element-wise arithmetic mean is then calculated from these 10,000 matrices to obtain a global average covariance matrix that represents the recent average operational statistics of the entire power grid.

[0070] After obtaining the global average covariance matrix, eigenvalue decomposition is performed on the matrix. This process produces a series of eigenvalues ​​and corresponding eigenvectors, where the eigenvectors represent the most predominant directions of data variation. All eigenvalues ​​are sorted from largest to smallest, and the 10 largest eigenvalues ​​are selected. The 10 eigenvectors corresponding to these 10 largest eigenvalues ​​are also selected, forming a new set of basis vectors. The updated basis vectors represent the most prevalent fluctuation patterns and correlation characteristics under the current power grid operating conditions.

[0071] In an optional embodiment, the Wasserstein distance of the low-dimensional state description sequences between nodes is calculated, and combined with the communication link quality, a weighted transition matrix for parameter aggregation between nodes is generated, including:

[0072] Get the packet loss rate of bidirectional communication between node i and its neighbor node j. and average round-trip delay Calculate the communication link quality factor ;

[0073] Obtain the Wasserstein distance between the low-dimensional state description subsequences of node i and its neighbor node j. ;

[0074] According to the formula Calculate the original weight of neighbor node j relative to node i;

[0075] Normalize the original weights of all neighboring nodes of node i to obtain the elements in the i-th row of the weighted transition matrix.

[0076] For example, for node i and its neighbor node j, the detected packet loss rate between them is 3%, that is... The value is 0.03, and the average round-trip time of the data packet is 20ms, that is... The value is 0.02s. The communication link quality factor is calculated according to the formula. The value is approximately 0.951. The difference in the operating states of these two nodes, i.e., the Wasserstein distance, was also calculated. The value is 0.4, indicating that there is a certain difference in the states of the two nodes.

[0077] Based on the above data, the original influence weight of neighbor node j on node i is calculated using a formula. Calculating e to the power of -0.4 yields approximately 0.670, which is then multiplied by the communication quality factor of 0.951, resulting in an original weight of approximately 0.637. Node i repeats this process for all neighbor nodes, such as k and l, obtaining their respective original weights, assumed to be 0.412 and 0.855. To ensure the total influence of all neighbor nodes is 1, normalization is required. The three original weights, 0.637, 0.412, and 0.855, are summed to obtain a total of 1.904. The normalized weights of nodes j, k, and l on node i are 0.637 / 1.904, 0.412 / 1.904, and 0.855 / 1.904, respectively. These values ​​form the row in the weighted transition matrix corresponding to node i.

[0078] S2, if the condition number of the local state covariance matrix of a node exceeds the first threshold, local optimization is initiated: the control parameters of neighboring nodes are aggregated using the weighted transition matrix, and the node's control parameters are updated through exponential mapping in combination with its own Riemann gradient; if the mean Wasserstein distance between global nodes exceeds the second threshold, global correction is initiated: the sparse optimization model is solved by the node with the highest betweenness centrality in the network to obtain global correction parameters, which are then distributed to each node to adjust the node's control parameters.

[0079] The condition number of the second-order state tensor of each node is continuously or periodically monitored. When the condition number exceeds a preset stability margin threshold, it indicates that the node state is becoming unstable. At this time, node i uses the weights in the weighted transition matrix to perform weighted aggregation on the control parameters of all neighboring nodes j. The aggregation operation is performed on the manifold, that is, first, the control parameters of neighboring nodes are mapped to the tangent space of node i through a logarithmic mapping, and then a weighted sum is performed to obtain the aggregated update direction. At the same time, the Riemann gradient of the control parameters is calculated according to the local objective function, such as minimizing power loss. The update direction is a linear combination of the aggregation direction and the negative Riemann gradient direction. The current control parameter is moved a small step along this final direction on the manifold through an exponential mapping to complete the parameter update.

[0080] Similarly, calculate the Wasserstein distance between all pairs of nodes in the network and find the arithmetic mean of the distances. When the arithmetic mean exceeds the second threshold representing the overall coordination of the system, it indicates that the global state of the system is diverging. Then, calculate the betweenness centrality of all nodes based on the network topology, i.e., the number of shortest paths through a node, and select the node with the highest centrality as the temporary central controller. The betweenness centrality can be calculated using networkx.betweenness_centrality(G), where G is an undirected graph of the network topology. Its calculation is existing technology, and the specific calculation steps will not be elaborated here.

[0081] The central node establishes a sparse optimization model with an L1 norm regularization term, aiming to minimize global operating costs and stability expenses. Solving the sparse optimization model yields a sparse global correction parameter vector, with most elements being zero. The central node broadcasts this global correction parameter vector to the entire network, and each node adds its control parameters to the corresponding components in the global correction parameter vector, completing a fast global correction.

[0082] In an optional embodiment, if the condition number of the local state covariance matrix of a node exceeds a first threshold, local optimization is initiated: the control parameters of neighboring nodes are aggregated using a weighted transition matrix, and the node's control parameters are updated via an exponential mapping, combining the node's own Riemann gradient, including:

[0083] When the condition number of the local state covariance matrix of node i exceeds the first threshold, the node performs the following operation:

[0084] The aggregation parameters are obtained by aggregating the control parameters of all neighboring nodes of a node using a weighted transition matrix. ;

[0085] Calculate its own control parameters Riemann gradient ;

[0086] Formulas combining exponential and logarithmic mappings:

[0087] Update the control parameters, where α is the learning rate, and... For exponential mapping, For logarithmic mapping, These are the updated control parameters.

[0088] The operating data of a generator node i in the power grid is analyzed to calculate the condition number of the local state covariance matrix of the generator node. The condition number is an indicator of the ill-conditioned nature of the matrix; an excessively high value usually indicates high correlation or potential instability in the data. Preferably, the first threshold is 1000. In a certain detection, the condition number of the generator node was found to be 1780, exceeding the preset threshold of 1000. This indicates that the generator node may be in an unstable state, triggering a local optimization program for the node.

[0089] After the local optimization program starts, the generator node uses the previously generated weighted transition matrix to aggregate the control parameters of all its neighboring nodes, forming a weighted aggregate parameter. The aggregation parameter represents the consensus state of the neighboring nodes. Specifically, all neighboring nodes N(i) directly connected to the generator node in the network topology are determined, and their current control parameters are obtained from all neighboring nodes j∈N(i). In one embodiment It is a matrix, obtained through the formula The calculation yielded that, For weights.

[0090] Generator nodes calculate their current control parameters gradient on the Stiefel manifold The gradient indicates the direction in which parameter adjustments improve the state most quickly. Generator nodes adjust their control parameters according to a given update formula. This update formula not only corrects along the negative gradient direction but also introduces a parameter that aligns the direction towards neighbor consensus. The corrective force is brought closer together. Setting the learning rate to 0.01 ensures that the step size of each adjustment is small, thereby updating the control parameters of generator node i to a more optimal position and reducing operational instability.

[0091] In an optional embodiment, if the mean Wasserstein distance between global nodes exceeds a second threshold, a global correction is initiated: the node with the highest network betweenness centrality solves the sparse optimization model to obtain global correction parameters, which are then distributed to each node to adjust the node's control parameters, including:

[0092] When the arithmetic mean of the Wasserstein distances between global nodes exceeds the second threshold, perform the following operation:

[0093] Calculate the betweenness centrality of all nodes in the network topology, and select the node with the highest betweenness centrality as the center node;

[0094] The central node constructs a LASSO regression model: Where A is the global state feature matrix, b is the ideal state vector, x is the global correction parameter to be solved, and λ is the regularization coefficient. It is the square L2 norm. It is an L1 norm;

[0095] Solve the LASSO regression model to obtain sparse global correction parameters x, and then distribute the global correction parameters.

[0096] The control center continuously monitors the differences in operating states between all node pairs in the entire power grid, i.e., the Wasserstein distance, and calculates the arithmetic mean of these distances. Preferably, the second threshold is 0.8; however, those skilled in the art should understand that the aforementioned first and second thresholds are merely example data and are not limited to the preferred values. When the calculated global average distance reaches 0.81 at a certain moment, exceeding the second threshold, it indicates a significant decrease in the consistency of the entire network's operating state, potentially indicating a systemic risk. At this point, the global correction mechanism is activated. The power grid topology is analyzed, the betweenness centrality of each node is calculated, and the node with the highest betweenness centrality value, such as a regional dispatch center, is selected as the central node for executing the global correction task.

[0097] The central node immediately collects the state data of all nodes in the entire network, forming a global state feature matrix A, and defines a target vector b representing the ideal operating state of the power grid. Using this data, the central node establishes a LASSO regression model. The goal of the LASSO regression model is to find a global correction parameter x such that the difference between the system state Ax after correction by the global correction parameter x and the ideal state b is minimized. A regularization term is used to constrain the number of non-zero elements in the global correction parameter x, with a regularization coefficient λ of 0.1. Solving the model yields a sparse global correction parameter x, where most elements are zero and only a few are non-zero. This means the correction scheme is highly targeted, requiring adjustment of only a few key parameters of critical nodes. The central node distributes the non-zero global correction parameters to the corresponding nodes for execution, achieving rapid global state correction of the entire power grid at minimal cost.

[0098] S3 decodes the updated control parameters into specific control instructions and executes them.

[0099] The control parameters, in matrix form and located on the manifold, are converted into physical commands executable by the power equipment using preset mapping rules. For example, the first element on the diagonal of the control parameter matrix might correspond to the active power setpoint of a generator set, and the second element to the reactive power setpoint; while the off-diagonal elements might correspond to the tap position of a transformer or the switching status of a capacitor bank. These specific numerical values ​​or status commands, after decoding, are sent to the local controller of the corresponding equipment, which then performs the physical operation.

[0100] An embodiment of the Internet of Things-based power equipment control system provided by this invention includes the following modules:

[0101] The projection and aggregation module is used to acquire real-time operating data of nodes of various power devices in the Internet of Things, and construct a second-order state tensor representing the operating state of each node; project the second-order state tensor onto a low-dimensional subspace defined by basis vectors to obtain a low-dimensional state descriptor, and the basis vectors are adaptively updated according to the historical state covariance matrix of all nodes; calculate the Wasserstein distance of the low-dimensional state descriptor sequences between nodes, and generate a weighted transition matrix for parameter aggregation between nodes by combining the communication link quality;

[0102] The correction module is used to initiate local optimization if the condition number of the local state covariance matrix of a node exceeds the first threshold: it aggregates the control parameters of neighboring nodes using a weighted transition matrix, combines its own Riemann gradient, and updates the node's control parameters through exponential mapping; if the mean Wasserstein distance between global nodes exceeds the second threshold, it initiates global correction: the node with the highest betweenness centrality in the network solves the sparse optimization model to obtain global correction parameters, which are then distributed to each node to adjust the node's control parameters.

[0103] The distribution module is used to decode the updated control parameters into specific control commands and execute them.

[0104] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A power equipment control method based on the Internet of Things, characterized in that, Includes the following steps: S1. Obtain real-time operating data of each power device node in the Internet of Things, and construct a second-order state tensor representing the operating state of each node; project the second-order state tensor onto a low-dimensional subspace defined by basis vectors to obtain a low-dimensional state descriptor, and adaptively update the basis vectors according to the historical state covariance matrix of all nodes; calculate the Wasserstein distance between the low-dimensional state descriptor sequences of nodes, and generate a weighted transition matrix for parameter aggregation between nodes by combining the communication link quality. S2, if the condition number of the local state covariance matrix of a node exceeds the first threshold, local optimization is initiated: the control parameters of neighboring nodes are aggregated using the weighted transition matrix, and the node's control parameters are updated through exponential mapping in combination with its own Riemann gradient; if the mean Wasserstein distance between global nodes exceeds the second threshold, global correction is initiated: the sparse optimization model is solved by the node with the highest betweenness centrality in the network to obtain global correction parameters, which are then distributed to each node to adjust the node's control parameters. S3 decodes the updated control parameters into specific control instructions and executes them.

2. The power equipment control method based on the Internet of Things according to claim 1, characterized in that, Acquire real-time operating data of nodes in the Internet of Things (IoT) for each power device, and construct a second-order state tensor representing the operating state of each node, including: The voltage U, current I, frequency f, active power P, and reactive power Q of each power equipment node are collected as a real-time operating data sequence over 50 consecutive sampling time points. The data sequence of the five physical quantities collected is used as the five-dimensional state vector of the node; Calculate the covariance matrix of the five-dimensional state vector of the node at 50 sampling time points, and use the covariance matrix as a second-order state tensor.

3. The power equipment control method based on the Internet of Things according to claim 1, characterized in that, The basis vectors are adaptively updated based on the historical state covariance matrix of all nodes, including: Collect the historical state covariance matrices of all nodes over the past 100 periods, calculate the arithmetic mean of the historical state covariance matrices, and obtain the global average covariance matrix. Perform eigenvalue decomposition on the global average covariance matrix; Select the eigenvectors corresponding to the top 10 largest eigenvalues ​​as the updated basis vectors.

4. The power equipment control method based on the Internet of Things according to claim 1, characterized in that, Calculate the Wasserstein distance between the low-dimensional state description sequences of the nodes, and combine this with the communication link quality to generate a weighted transition matrix for parameter aggregation between nodes, including: Get the packet loss rate of bidirectional communication between node i and its neighbor node j. and average round-trip delay Calculate the communication link quality factor. ; Obtain the Wasserstein distance between the low-dimensional state description subsequences of node i and its neighbor node j. ; According to the formula Calculate the original weights of neighbor node j with respect to node i. ; Normalize the original weights of all neighboring nodes of node i to obtain the elements in the i-th row of the weighted transition matrix.

5. The power equipment control method based on the Internet of Things according to claim 1, characterized in that, If the condition number of a node's local state covariance matrix exceeds a first threshold, local optimization is initiated: the control parameters of neighboring nodes are aggregated using a weighted transition matrix, and the node's control parameters are updated via an exponential mapping, combining the node's own Riemann gradient. This includes: When the condition number of the local state covariance matrix of node i exceeds the first threshold, the node performs the following operation: The aggregation parameters are obtained by aggregating the control parameters of all neighboring nodes of a node using a weighted transition matrix. ; Calculate its own control parameters Riemann gradient ; Formulas combining exponential and logarithmic mappings: Update the control parameters, where α is the learning rate, and... For exponential mapping, For logarithmic mapping, These are the updated control parameters.

6. The power equipment control method based on the Internet of Things according to any one of claims 1-5, characterized in that, If the mean Wasserstein distance between global nodes exceeds the second threshold, a global correction is initiated: the node with the highest betweenness centrality solves the sparse optimization model to obtain global correction parameters, which are then distributed to each node to adjust the node's control parameters, including: When the arithmetic mean of the Wasserstein distances between global nodes exceeds the second threshold, perform the following operation: Calculate the betweenness centrality of all nodes in the network topology, and select the node with the highest betweenness centrality as the center node; The central node constructs a LASSO regression model: , where A is the global state feature matrix, b is the ideal state vector, x is the global correction parameter to be solved, and λ is the regularization coefficient; Solve the LASSO regression model to obtain sparse global correction parameters x, and then distribute the global correction parameters.

7. A power equipment control system based on the Internet of Things, characterized in that, Includes the following modules: The projection and aggregation module is used to acquire real-time operating data of nodes of various power devices in the Internet of Things, and construct a second-order state tensor representing the operating state of each node; project the second-order state tensor onto a low-dimensional subspace defined by basis vectors to obtain a low-dimensional state descriptor, and the basis vectors are adaptively updated according to the historical state covariance matrix of all nodes; calculate the Wasserstein distance of the low-dimensional state descriptor sequences between nodes, and generate a weighted transition matrix for parameter aggregation between nodes by combining the communication link quality; The correction module is used to initiate local optimization if the condition number of the local state covariance matrix of a node exceeds the first threshold: it aggregates the control parameters of neighboring nodes using a weighted transition matrix, combines its own Riemann gradient, and updates the node's control parameters through exponential mapping; if the mean Wasserstein distance between global nodes exceeds the second threshold, it initiates global correction: the node with the highest betweenness centrality in the network solves the sparse optimization model to obtain global correction parameters, which are then distributed to each node to adjust the node's control parameters. The distribution module is used to decode the updated control parameters into specific control commands and execute them.

8. The Internet of Things-based power equipment control system according to claim 7, characterized in that, Acquire real-time operating data of nodes in the Internet of Things (IoT) for each power device, and construct a second-order state tensor representing the operating state of each node, including: The voltage U, current I, frequency f, active power P, and reactive power Q of each power equipment node are collected as a real-time operating data sequence over 50 consecutive sampling time points. The data sequence of the five physical quantities collected is used as the five-dimensional state vector of the node; Calculate the covariance matrix of the five-dimensional state vector of the node at 50 sampling time points, and use the covariance matrix as a second-order state tensor.

9. The Internet of Things-based power equipment control system according to claim 7, characterized in that, The basis vectors are adaptively updated based on the historical state covariance matrix of all nodes, including: Collect the historical state covariance matrices of all nodes over the past 100 periods, calculate the arithmetic mean of the historical state covariance matrices, and obtain the global average covariance matrix. Perform eigenvalue decomposition on the global average covariance matrix; Select the eigenvectors corresponding to the top 10 largest eigenvalues ​​as the updated basis vectors.

10. The Internet of Things-based power equipment control system according to claim 7, characterized in that, Calculate the Wasserstein distance between the low-dimensional state description sequences of the nodes, and combine this with the communication link quality to generate a weighted transition matrix for parameter aggregation between nodes, including: Get the packet loss rate of bidirectional communication between node i and its neighbor node j. and average round-trip delay Calculate the communication link quality factor. ; Obtain the Wasserstein distance between the low-dimensional state description subsequences of node i and its neighbor node j. ; According to the formula Calculate the original weights of neighbor node j with respect to node i. ; Normalize the original weights of all neighboring nodes of node i to obtain the elements in the i-th row of the weighted transition matrix.

Images