Full-dimensional Real-time Monitoring and Stable Operation Linkage System for Power System

Abstract

The present invention relates to the technical field of power system monitoring and control, and discloses a full-dimensional real-time monitoring and stable operation linkage system for a power system. The system includes a multi-dimensional data acquisition module that acquires and cleans operation data; a spatio-temporal feature mapping module that generates a spatio-temporal feature coordinate cluster containing multi-dimensional correlation relationships; a dynamic monitoring module that adjusts thresholds, divides monitoring intervals, and extracts indicators; a stable operation linkage control module that constructs a model to generate a steady-state operation control strategy; and a feedback optimization module that fuses multi-source data to output regulation instructions. The system also has a warning module that performs correlation analysis on real-time regulation instructions and equipment health and triggers an emergency response. Through the collaborative work of each module, full-dimensional real-time monitoring and stable operation control of the power system are achieved, the data processing and analysis capabilities and the precise regulation level are improved, the system stability and reliability are enhanced, and the equipment loss and operation cost are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of power system monitoring and control, and in particular to a full-dimensional real-time monitoring and stable operation linkage system for a power system. Background Art

[0002] As modern society places increasing demands on the stability and reliability of power supply, and the scale and complexity of power systems continue to increase, traditional monitoring and control technologies face many challenges.

[0003] In terms of data collection, early power system monitoring only focused on some key equipment or a few operating parameters. The data dimension was single, and it was difficult to fully reflect the overall operating status of the system. For example, only the voltage and current of the substation were monitored, ignoring the environmental parameters such as temperature and humidity of the equipment and the correlation information between the equipment, and it was impossible to detect potential fault hazards in time. Even though the data collection dimensions of some systems have increased today, the collection frequency is low and the real-time performance of the data is poor, which cannot meet the rapidly changing monitoring needs of the power system. When responding to emergencies, it is difficult to make accurate decisions in a timely manner due to the inability to obtain real-time data, resulting in problems such as the expansion of power outages and prolonged power supply restoration time.

[0004] For data processing and analysis, traditional methods rely on simple threshold judgments and manual experience analysis. Faced with massive and complex power data, this method is inefficient and prone to misjudgment. For example, when multiple operating parameters change slightly at the same time, a single threshold judgment cannot accurately determine whether the system is in an abnormal state. Manual analysis is not only time-consuming and labor-intensive, but may also affect the accuracy of judgment due to subjective factors. Moreover, traditional methods are difficult to explore the deep correlation between data and cannot provide strong support for the optimized operation of the power system.

[0005] In terms of the coordination of monitoring and control, traditional system monitoring and control are independent of each other and lack an effective linkage mechanism. When an abnormality is detected, the information cannot be transmitted to the control system in a timely manner and corresponding adjustments cannot be made, resulting in the expansion of the impact of the fault. For example, when a transmission line is overloaded, although the monitoring system can detect the problem, the control measures cannot respond quickly, which may cause the line to burn out and affect the power supply of the entire area. In addition, traditional control strategies are often based on fixed models and preset rules, and cannot be dynamically adjusted according to real-time operating conditions, making it difficult to adapt to the complex and changeable operating environment of the power system.

[0006] At the same time, the operation of the power system is affected by many external factors, such as meteorological conditions and changes in user load. Traditional monitoring and control systems do not take these external factors into account and are unable to include them in the scope of comprehensive analysis and control. During peak periods of electricity consumption such as hot weather or holidays, the substantial increase in user load and the impact of meteorological factors on equipment operation are not fully considered, which can easily lead to increased pressure on the operation of the power system and even failures. Summary of the Invention

[0007] The object of the present invention is to provide a full-dimensional real-time monitoring and stable operation linkage system for a power system to solve the problems raised in the above-mentioned background technology.

[0008] To achieve the above object, the present invention provides the following technical solution: A full-dimensional real-time monitoring and stable operation linkage system for a power system, the system includes:

[0009] A multi-dimensional data acquisition module, which is used to obtain an operation data set within a preset time range in the power system, and preprocess the operation data set according to a preset multi-dimensional data cleaning strategy to generate standardized operation data;

[0010] A spatio-temporal feature mapping module, which is used to map the standardized operation data into a spatio-temporal feature space, and generate a spatio-temporal feature coordinate cluster based on a multi-dimensional spatio-temporal clustering algorithm. The spatio-temporal feature coordinate cluster includes multi-dimensional correlation relationships of equipment status, load distribution, and environmental parameters;

[0011] A dynamic monitoring module, which is used to adaptively adjust the threshold of the spatio-temporal feature coordinate cluster according to a preset anomaly detection rule, divide a real-time monitoring interval through a mixed integer linear programming algorithm, and extract an operation index set within each interval;

[0012] A stable operation linkage control module, which is used to construct a linkage decision-making unit including a dynamic topology optimization network model, and use the operation index set to iteratively update the parameters of the dynamic topology optimization network model to generate a steady-state operation control strategy;

[0013] A feedback optimization module, which is used to perform multi-source data fusion optimization on the steady-state operation control strategy according to a preset global sensitivity analysis model, and output a full-dimensional real-time regulation instruction for the power system.

[0014] Preferably, generating the spatio-temporal feature coordinate cluster based on the multi-dimensional spatio-temporal clustering algorithm includes:

[0015] Extracting timestamps, device IDs, and operation parameters in the standardized operation data to construct a multi-dimensional spatio-temporal matrix;

[0016] Using a sliding window segmentation algorithm to segment the multi-dimensional spatio-temporal matrix to generate a set of time series sub-matrices;

[0017] Adopting a density-based clustering model to perform spatial density analysis on the set of time series sub-matrices to eliminate isolated noise points;

[0018] Performing multi-level aggregation on the remaining sub-matrices through a hierarchical clustering algorithm to generate a feature coordinate cluster with spatio-temporal correlation.

[0019] Preferably, the adaptive threshold adjustment includes:

[0020] Calculating the mean and variance of key operating indicators in each interval according to the historical operating data distribution in the real-time monitoring interval;

[0021] Based on the dynamic sliding window algorithm, exponentially smoothing and correcting the mean and variance to generate a dynamic threshold benchmark;

[0022] Using fuzzy logic rules to blur the interval boundaries of the dynamic threshold benchmark to generate an adaptive threshold interval.

[0023] Preferably, the hybrid integer linear programming algorithm for dividing the real-time monitoring interval includes:

[0024] Defining the objective function as the weighted minimization of load fluctuation and equipment loss in the monitoring interval;

[0025] Setting the constraint conditions as the grid topology connectivity, equipment capacity limit, and environmental parameter tolerance range;

[0026] Solving the objective function by the branch and bound method and outputting the optimal monitoring interval division scheme.

[0027] Preferably, the parameter iterative update of the dynamic topology optimization network model includes:

[0028] Inputting the set of operating indicators into the input layer of the dynamic topology optimization network model, and using the graph convolutional network to extract the topological feature vector;

[0029] Assigning weights to the topological feature vector through the attention mechanism to generate a device priority sequence;

[0030] Adopting the backpropagation algorithm combined with the genetic algorithm to optimize the network weights and update the decision parameters of the linkage decision unit.

[0031] Preferably, the multi-dimensional data cleaning strategy includes:

[0032] Identifying duplicate data segments in the set of operating data and removing duplicates based on the timestamp alignment rule;

[0033] Detecting missing data points and interpolating the missing data points using the spatio-temporal Kriging interpolation method;

[0034] Normalizing the interpolated data to generate normalized operating data with a mean of zero and a variance of one.

[0035] Preferably, the global sensitivity analysis model includes:

[0036] Construct a multi-variable sensitivity calculation framework based on the Sobol index method to quantify the sensitivity of control instructions to each operating parameter;

[0037] Generate a parameter perturbation sample set through Monte Carlo sampling and calculate the variance contribution rate of each parameter;

[0038] Screen the parameters with a contribution rate higher than the preset threshold as the core variables for optimized control.

[0039] Preferably, the multi-source data fusion optimization includes:

[0040] Integrate device sensor data, meteorological data, and user load prediction data to construct a heterogeneous data fusion matrix;

[0041] Use the tensor decomposition algorithm to perform dimensionality reduction processing on the heterogeneous data fusion matrix and extract the principal component eigenvectors;

[0042] Input the principal component eigenvectors into the feedback optimization module to generate optimized control instructions.

[0043] Preferably, the system further includes:

[0044] Construct an early warning module including a device health assessment function, perform correlation analysis on the real-time control instructions and the device health, and output a hierarchical early warning signal;

[0045] Trigger a preset emergency response mechanism according to the hierarchical early warning signal to generate device maintenance or load adjustment instructions.

[0046] Preferably, the present invention further includes an electronic device, and the device further includes:

[0047] At least one processor; and

[0048] A memory communicatively connected to the at least one processor; wherein,

[0049] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the operations of the above-mentioned full-dimensional real-time monitoring and stable operation linkage system of the power system.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] The full-dimensional real-time monitoring and stable operation linkage system of the power system of the present invention has many remarkable beneficial effects. At the level of data processing and analysis, the multi-dimensional data acquisition module can obtain the operation data set of the power system within a preset time range, and perform preprocessing through the multi-dimensional data cleaning strategy to generate standardized operation data, greatly improving the data quality. The spatio-temporal feature mapping module maps the standardized data to the spatio-temporal feature space, and generates a spatio-temporal feature coordinate cluster containing multi-dimensional correlation relationships of equipment status, load distribution, and environmental parameters by means of a multi-dimensional spatio-temporal clustering algorithm, deeply mining the potential connections between data. This can not only grasp the operation status of the power system more accurately, but also detect abnormal signs in advance. For example, by analyzing the correlation between equipment status and environmental parameters, the failure risk of equipment in a specific environment can be predicted. Compared with traditional single-parameter monitoring, the failure prediction accuracy is greatly improved, effectively reducing the incidence of power outages caused by sudden failures.

[0052] The dynamic monitoring module adaptively adjusts the threshold for the spatio-temporal feature coordinate cluster according to the preset anomaly detection rules, and uses the mixed-integer linear programming algorithm to divide the real-time monitoring interval and extract the operation index set. The adaptive threshold adjustment can dynamically change according to the historical operation data of the real-time monitoring interval. Compared with traditional fixed-threshold monitoring, the judgment of abnormal situations is more accurate and sensitive, reducing false alarms and missed alarms. The mixed-integer linear programming algorithm divides the monitoring interval, comprehensively considering factors such as load fluctuations, equipment losses, grid topological connectivity, equipment capacity limits, and environmental parameter tolerance ranges, making the monitoring interval division more scientific and reasonable, thereby improving the monitoring efficiency, timely discovering potential safety hazards, and ensuring the safe and stable operation of the power system.

[0053] In terms of operation control, the stable operation linkage control module constructs a linkage decision-making unit containing a dynamic topology optimization network model, and uses the operation index set to iteratively update the parameters of the model to generate a steady-state operation control strategy. By extracting the topological feature vector through the graph convolutional network, combining the attention mechanism to generate the equipment priority sequence, and then using the backpropagation algorithm combined with the genetic algorithm to optimize the network weights, this method can dynamically adjust the control strategy according to the real-time operation conditions, realizing the precise regulation of the power system. When the load changes, the power distribution can be quickly adjusted to ensure the normal operation of each device, reduce equipment losses, and improve the operation efficiency of the power system. Compared with traditional fixed control strategies, it can effectively reduce energy waste and operating costs.

[0054] The feedback optimization module optimizes the steady-state operation control strategy through multi-source data fusion according to a preset global sensitivity analysis model, and outputs full-dimensional real-time regulation instructions. This module integrates device sensor data, meteorological data, and user load prediction data, and extracts the principal component feature vectors through dimensionality reduction processing using the tensor decomposition algorithm, making the regulation instructions more scientific and forward-looking. Considering that meteorological data can anticipate the impact of severe weather on the power system, and combining user load prediction data can optimize power dispatching to avoid power supply-demand imbalance, further enhancing the stability and reliability of the power system.

[0055] In addition, the system is also equipped with an early warning module that constructs an equipment health assessment function, correlates the real-time regulation instructions with the equipment health, outputs graded early warning signals, and triggers an emergency response mechanism to generate equipment maintenance or load adjustment instructions. This can promptly warn of potential equipment failures, arrange maintenance in advance, extend the service life of the equipment, and at the same time reasonably adjust the load when the equipment has problems, ensuring uninterrupted power supply and improving the user's electricity consumption experience. Brief Description of the Drawings

[0056] Figure 1 is the working principle diagram of the full-dimensional real-time monitoring and stable operation linkage system of the power system described in the present invention;

[0057] Figure 2 is the working principle diagram of the adaptive threshold adjustment;

[0058] Figure 3 is the working principle diagram of the multi-dimensional data cleaning strategy;

[0059] Figure 4 is the working principle diagram of the early warning and emergency response. Detailed Embodiments

[0060] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0061] Please refer to Figures 1 - 4 , the present invention relates to a full-dimensional real-time monitoring and stable operation linkage system for a power system, aiming to achieve all-round and real-time monitoring and stable operation control of the power system. The overall implementation plan is as follows:

[0062] Multi-dimensional Data Acquisition Module: During the operation of the power system, the multi-dimensional data acquisition module obtains operation data according to a preset time range. This preset time range can be flexibly set according to actual needs, such as in minutes, hours, etc. The collected data constitutes an operation data set, covering information in many aspects of the power system, such as equipment operation status data, load data, environmental data, etc. After obtaining the data, the module preprocesses the operation data set according to the preset multi-dimensional data cleaning strategy. By identifying duplicate data segments and de-duplicating them according to the time stamp alignment rule, detecting missing data points and interpolating them using the spatio-temporal Kriging interpolation method, and finally standardizing the interpolated data, standardized operation data is finally generated, providing a high-quality data basis for the work of subsequent modules.

[0063] Spatio-temporal Feature Mapping Module: This module receives the standardized operation data generated by the multi-dimensional data acquisition module and maps it into the spatio-temporal feature space. Specifically, first extract the time stamp, equipment ID, and operation parameters in the standardized operation data to construct a multi-dimensional spatio-temporal matrix. Then use the sliding window segmentation algorithm to segment the multi-dimensional spatio-temporal matrix to form a set of time series sub-matrices. Next, use a density-based clustering model to perform spatial density analysis on the set of time series sub-matrices, eliminate isolated noise points, and exclude the interference of abnormal data. Finally, use the hierarchical clustering algorithm to perform multi-level aggregation on the remaining sub-matrices, thereby generating a spatio-temporal feature coordinate cluster containing multi-dimensional correlation relationships of equipment status, load distribution, and environmental parameters, and mining the potential spatio-temporal correlation characteristics between data.

[0064] Dynamic Monitoring Module: The dynamic monitoring module adaptively adjusts the threshold for the spatio-temporal feature coordinate cluster according to the preset anomaly detection rules. According to the historical operation data distribution in the real-time monitoring interval, calculate the mean and variance of the key operation indicators in each interval; then based on the dynamic sliding window algorithm, perform exponential smoothing correction on the mean and variance to generate a dynamic threshold benchmark; then use fuzzy logic rules to blur the interval boundaries of the dynamic threshold benchmark to obtain an adaptive threshold interval. At the same time, divide the real-time monitoring interval through the mixed integer linear programming algorithm. Specifically, define the objective function as the weighted minimization of the load fluctuation and equipment loss in the monitoring interval, set the constraint conditions as the grid topology connectivity, equipment capacity limit, and environmental parameter tolerance range, solve the objective function through the branch and bound method, output the optimal monitoring interval division scheme, and then extract the operation index set in each interval to provide key data for subsequent analysis.

[0065] Steady operation linkage control module: The steady operation linkage control module constructs a linkage decision-making unit containing a dynamic topology optimization network model. The set of operation indicators extracted by the dynamic monitoring module is input into the input layer of the dynamic topology optimization network model, and the graph convolutional network is used to extract the topology feature vector; the attention mechanism is used to assign weights to the topology feature vector to generate the device priority sequence; the backpropagation algorithm combined with the genetic algorithm is used to optimize the network weights and update the decision parameters of the linkage decision-making unit, thereby generating a steady-state operation control strategy to achieve effective regulation of the power system operation.

[0066] Feedback optimization module: The feedback optimization module performs multi-source data fusion optimization on the steady-state operation control strategy according to the preset global sensitivity analysis model. A multi-variable sensitivity calculation framework based on the Sobol index method is constructed to quantify the sensitivity of the regulation instructions to each operation parameter; the parameter perturbation sample set is generated through Monte Carlo sampling, and the variance contribution rate of each parameter is calculated; the parameters with a contribution rate higher than the preset threshold are selected as the core variables for optimized regulation. At the same time, the device sensor data, meteorological data, and user load prediction data are integrated to construct a heterogeneous data fusion matrix; the tensor decomposition algorithm is used to perform dimensionality reduction processing on the heterogeneous data fusion matrix to extract the principal component feature vector; the principal component feature vector is input into the feedback optimization module to generate full-dimensional real-time regulation instructions for the power system, realizing precise optimization control of the power system operation.

[0067] The following further illustrates the implementation of the present invention in combination with Embodiments 1 to 6.

[0068] Embodiment 1:

[0069] When the power system is operating, the multi-dimensional data acquisition module starts to work first. For the set of operation data collected, it contains a large amount of operation data from different devices and different times. When identifying duplicate data segments, the duplicate data part is determined by comparing the timestamps and other key identification information in the data. For example, if there are two records with exactly the same data except for the timestamps, and the timestamps are within the error range allowed by the timestamp alignment rule, they are considered duplicate data segments and de-duplication operations are performed. For the detected missing data points, the spatio-temporal Kriging interpolation method is used for interpolation. The spatio-temporal Kriging interpolation method is based on the principle of spatial autocorrelation and uses the information of surrounding known data points to estimate the value of the missing point. Suppose there are multiple devices collecting data in a region, the position coordinates of each device are , and the corresponding operation parameter values are . For a certain missing data point , by calculating the spatial distance between the surrounding data points and the missing point, as well as the spatial autocorrelation function, the weights of each known data point for the estimated value of the missing point are determined, and then the interpolation value of the missing point is obtained.

[0070] After completing data imputation, the data is standardized so that its mean is zero and variance is one. Let the original data be , and the standardized data be . The standardization formula is , where is the mean of the original data, and is the standard deviation of the original data. After standardization, standardized operation data is generated.

[0071] After receiving the standardized operation data, the spatio-temporal feature mapping module extracts the timestamps, device IDs, and operation parameters therein to construct a multi-dimensional spatio-temporal matrix. For example, with time as rows, device IDs as columns, and operation parameters as matrix elements, a multi-dimensional matrix structure is constructed. Then, the sliding window segmentation algorithm is used to segment the multi-dimensional spatio-temporal matrix. Next, a density-based clustering model is used to perform spatial density analysis on the set of time series sub-matrices. This model calculates the density values around each sub-matrix by setting a density threshold, and identifies the sub-matrices with density values lower than the threshold as isolated noise points and eliminates them. Finally, hierarchical clustering algorithm is used to perform multi-level aggregation on the remaining sub-matrices. The hierarchical clustering algorithm starts with each sub-matrix as a separate class, and continuously merges similar classes according to the similarity measure (such as Euclidean distance, etc.) between sub-matrices, forming a feature coordinate cluster with spatio-temporal correlation, thus realizing the effective extraction of data spatio-temporal features.

[0072] Example 2:

[0073] In the dynamic monitoring link of the power system, adaptive threshold adjustment can enable the monitoring system to better adapt to the complex and changeable operating states of the power system.

[0074] During real-time monitoring, the system continuously records the historical operation data of each monitoring interval. Taking a specific monitoring interval as an example, assume that the key operation index in this interval is the power load value. Over a period of time, the system has collected a series of load data points, such as . First, calculate the mean of these data points. The calculation formula for the mean is , which reflects the average level of power load in this monitoring interval, where is the th load data point. At the same time, calculate the variance . The calculation formula for the variance is , and the variance reflects the degree of dispersion of the load data, that is, the magnitude of the load fluctuation.

[0075] Based on the dynamic sliding window algorithm, exponential smoothing correction is performed on the mean and variance. The dynamic sliding window continuously updates the included data over time. Assume the size of the sliding window is , at a certain moment, the latest load data included in the window is . When performing exponential smoothing correction, the calculation formula for the new mean is , where is the smoothing coefficient, with a value range between 0 and 1, is the mean of the data within the window at the previous moment. The calculation formula for the new variance is , is the variance of the data within the window at the previous moment. Through this exponential smoothing correction, the change trend of the load data can be reflected more timely, making the calculation results of the mean and variance more timely.

[0076] Using fuzzy logic rules to perform interval boundary fuzzification on the dynamic threshold benchmark to generate an adaptive threshold interval. Fuzzy logic rules are based on comprehensive judgment of multiple input variables. For example, the deviation between the current mean and the historical mean, the current variance, and the change trend of the recent load, etc. are used as input variables. Define fuzzy sets, such as dividing the load deviation into fuzzy categories such as "large negative", "small negative", "zero", "small positive", "large positive", etc., and dividing the variance into fuzzy categories such as "low", "medium", "high", etc. Then formulate a series of fuzzy inference rules, such as "if the load deviation is large positive and the variance is medium, then the threshold is appropriately increased". According to these rules, the upper and lower boundaries of the dynamic threshold benchmark are adjusted to more accurately judge whether the operation state of the power system is abnormal.

[0077] Example 3:

[0078] In the real-time monitoring of the power system, using the mixed-integer linear programming algorithm to divide the real-time monitoring interval can improve the accuracy and effectiveness of monitoring and ensure the stable operation of the power system.

[0079] When applying the mixed-integer linear programming algorithm, the objective function needs to be defined first. The objective function is set to minimize the weighted sum of the load fluctuation and the equipment loss within the monitoring interval. Assume that within a monitoring interval, there are moments, the corresponding load values are , the equipment loss values are , the weight of the load fluctuation is set to , and the weight of the equipment loss is . Then the expression of the objective function is . Among them, and is a weight coefficient set according to the actual operation conditions and requirements of the power system, used to balance the relative importance of load fluctuations and equipment losses in the optimization process. is the load value at the -th moment within the monitoring interval. is the load value at the -th moment within the monitoring interval. is the equipment loss value at the -th moment within the monitoring interval. For example, if the current power system pays more attention to the stable operation of equipment, the value of can be appropriately increased; if more attention is paid to the smooth change of load, the weight of can be increased.

[0080] Next, set the constraint conditions, mainly including the grid topology connectivity, equipment capacity limit, and environmental parameter tolerance range. The grid topology connectivity constraint ensures that when dividing the monitoring interval, the connection relationship between each part of the grid conforms to the actual grid structure, and unreasonable situations such as grid islands will not occur. In terms of equipment capacity limit, let the upper limit of the capacity of the -th equipment be . Within the monitoring interval, the load ( represents the moment) flowing through this equipment must satisfy . For example, the rated capacity of a certain transformer is . Throughout the monitoring interval, the load passing through this transformer at any moment cannot exceed this rated value. The environmental parameter tolerance range takes into account the influence of environmental factors such as temperature and humidity on the operation of the power system. Assume that the tolerance range of the environmental temperature is . Within the monitoring interval, the environmental temperature needs to satisfy . If the temperature exceeds this range, it may affect the performance and safety of the equipment, and then affect the normal operation of the power system.

[0081] Finally, the objective function is solved by the branch and bound method, and the optimal monitoring interval division scheme is output. The branch and bound method is a commonly used algorithm for solving integer programming problems. It first decomposes the original problem into multiple sub-problems, and solves each sub-problem to obtain a lower bound. Then, by continuously comparing the lower bounds of each sub-problem with the currently found optimal solution (upper bound), sub-problems that cannot produce the optimal solution are gradually excluded, narrowing the search scope. During the solution process, the optimal solution is continuously updated until the optimal monitoring interval division scheme that meets the conditions is found. For example, in actual calculations, a certain sub-problem may be solved first to obtain a preliminary division scheme, and the value of its objective function is calculated as the current upper bound. Then, other sub-problems are continued to be solved. If the lower bound of a certain sub-problem is greater than the current upper bound, this sub-problem and its branches can be discarded and no further in-depth calculation is required. In this way, the optimal monitoring interval division is finally obtained, so that the weighted sum of load fluctuations and equipment losses is minimized under the premise of meeting various constraints.

[0082] Embodiment 4:

[0083] In the stable operation linkage control module, parameter iterative update of the dynamic topology optimization network model is the core step to achieve precise control of the power system. Specifically, it includes:

[0084] Input the set of operation indicators into the input layer of the dynamic topology optimization network model. These sets of operation indicators cover data in multiple important aspects of the power system, such as power load, voltage, current, equipment temperature, etc. Taking a certain local power grid area as an example, there are multiple devices in this area, and each device generates different operation data at different times. After organizing these data into a format that meets the input requirements of the model, they are input into the model.

[0085] Use the graph convolutional network to extract topological feature vectors. Assume that the network topology structure of the power system can be represented as a graph , where represents the set of nodes, corresponding to each device in the power system; represents the set of edges, corresponding to the connection relationships between devices. The graph convolutional network can effectively extract the topological relationship features between nodes by performing convolutional operations on this graph structure. During the convolutional process, the convolutional kernel slides on the graph, fusing the feature information of each node and its neighbor nodes. For example, for a certain device node , its neighbor nodes are , and the graph convolutional network will comprehensively consider its own features and the features of its neighbor nodes, and generate the topological feature vector of this node through a series of calculations. These topological feature vectors reflect the position of the device in the entire power system topology structure and its association with other devices, providing an important basis for subsequent analysis.

[0086] Weight is assigned to the topological feature vectors through an attention mechanism to generate a device priority sequence. The attention mechanism can automatically assign weights according to the importance of different feature vectors. Suppose the generated topological feature vectors are , and the attention mechanism calculates the weight coefficient for each feature vector, such that , is the weight coefficient of the -th feature vector. The calculation of the weight coefficient is usually based on the similarity measure between feature vectors and some learnable parameters. For example, the cosine similarity between feature vectors is calculated to measure their similarity, and combined with some parameters learned during the training process, the weight of each feature vector is determined. According to the weight coefficient, the topological feature vectors are weighted and summed to obtain the comprehensive eigenvalue of each device. For example, the comprehensive eigenvalue of device . The devices are sorted according to the magnitude of the comprehensive eigenvalue to generate a device priority sequence. In this way, the devices that need to be focused on and regulated preferentially during the regulation process can be determined, improving the pertinence and effectiveness of the regulation.

[0087] The backpropagation algorithm is combined with the genetic algorithm to optimize the network weights and update the decision parameters of the linkage decision unit. The backpropagation algorithm calculates the error between the output result and the true value, propagates the error backward to each layer of the network, calculates the gradient of each layer's parameters, and adjusts the network weights according to the gradient descent method to gradually reduce the error. In the application scenario of the power system, the actual operating state of the power system is used as the true value, and the error is obtained by comparing the output result of the model with it. For example, the difference between the load value of a certain device predicted by the model and the actually monitored load value is the error. The network weights are continuously adjusted through the backpropagation algorithm to make the prediction result of the model closer to the true value. The genetic algorithm simulates the biological evolution process and searches for the optimal solution in the solution space through operations such as selection, crossover, and mutation. During the optimization process, the weight adjustment direction obtained by the backpropagation algorithm is used as the individual mutation direction in the genetic algorithm. Combining the global search ability of the genetic algorithm, the optimal network weights can be found more effectively. For example, in the selection operation of the genetic algorithm, individuals with high fitness (i.e., small error) are selected and retained for the next generation; in the crossover operation, part of the genes of two individuals are exchanged to generate new individuals; in the mutation operation, the genes of individuals are randomly changed with a certain probability. Through continuous iteration, the decision parameters of the linkage decision unit are finally updated to generate a more reasonable steady-state operation control strategy to ensure the stable operation of the power system.

[0088] Example 5:

[0089] In the feedback optimization module, a multi-variable sensitivity calculation framework based on the Sobol index method is constructed to quantify the sensitivity of control instructions to each operating parameter. Let the control instruction be , and the operating parameter be , and the model output be . The Sobol index method measures the parameter sensitivity by calculating the total effect index and the first-order effect index . The total effect index represents the total influence degree of all parameters on the output result, which reflects the interaction between parameters and the comprehensive influence of a single parameter on the output. The first-order effect index represents the direct influence degree of a single parameter on the output result. Calculating and requires generating a large number of sample points through Monte Carlo sampling. Monte Carlo sampling is a method based on random sampling. In the parameter space, a large number of sample points are randomly drawn according to a certain probability distribution (such as uniform distribution, etc.). For example, for the parameter , a series of values are randomly drawn within its value range, and at the same time, corresponding random sampling is also performed on other parameters to form sample points , . These sample points are substituted into the model to calculate the output value of each sample point, and then the sensitivity index of the parameter is obtained.

[0090] Generate a parameter perturbation sample set through Monte Carlo sampling and calculate the variance contribution rate of each parameter. When generating the parameter perturbation sample set, each parameter is randomly perturbed within a certain range. Suppose parameter is perturbed to generate a sample set , and calculate the model output value corresponding to each sample point. First, calculate the variance of the sample output value, , where is the mean value of the sample output value, and is the output value after the -th sample point is substituted into the model. Then calculate the variance component related to parameter , and obtain through a specific calculation method (based on the principle of the Sobol index method). Finally, the variance contribution rate of parameter is obtained. The larger the variance contribution rate , the more significant the parameter The greater the impact on the regulation command.

[0091] Select parameters with a contribution rate higher than the preset threshold as the core variables for optimized regulation. The preset threshold is set according to practical experience and system requirements. For example, set the threshold to 0.1. If the variance contribution rate of a certain parameter is such that, then it is determined as the core variable for optimized regulation. These core variables have a greater impact on the regulation command. In the subsequent optimization process, focusing on regulating and optimizing these variables can more effectively improve the operating performance of the power system. For example, if it is found that the load parameter of a certain device has a high variance contribution rate, when formulating a regulation strategy, the load of this device can be preferentially considered for adjustment to achieve a better optimization effect.

[0092] Example 6:

[0093] In the multi-source data fusion optimization process of the feedback optimization module, first perform multi-source data fusion optimization. Integrate device sensor data, meteorological data, and user load prediction data to construct a heterogeneous data fusion matrix. The device sensor can collect the operating status information of the device in real time, such as the temperature, pressure, current, etc. of the device; the meteorological data includes air temperature, humidity, wind speed, etc. These meteorological factors will affect the load and device operation of the power system. For example, high-temperature weather may lead to an increase in the air-conditioning load of residents; the user load prediction data is based on historical electricity consumption data and user behavior analysis to predict the electricity consumption demand of users in the future period. Arrange these different types of data according to certain rules to construct a multi-dimensional heterogeneous data fusion matrix. For example, using time as the row index, taking device sensor data, meteorological data, and user load prediction data as different columns respectively, and filling each data into the corresponding position to form a matrix structure containing multi-source information.

[0094] Use the tensor decomposition algorithm to perform dimensionality reduction processing on the heterogeneous data fusion matrix and extract the principal component eigenvectors. The tensor decomposition algorithm can decompose high-dimensional tensor data into a combination of multiple low-dimensional tensors. For the constructed heterogeneous data fusion matrix (which can be regarded as a tensor), through tensor decomposition, it is decomposed into a core tensor and multiple factor matrices. During the decomposition process, the algorithm will automatically extract the main feature information in the data, and this information is concentrated in the factor matrices. Extract the principal component eigenvectors among them. These vectors retain most of the key information of the original data, while reducing the dimension of the data and the amount of calculation. For example, after tensor decomposition, the originally high-dimensional data containing a large amount of redundant information is transformed into several principal component eigenvectors, which can more concisely represent the core features of the original data.

[0095] The principal component eigenvector is input into the feedback optimization module to generate optimized control instructions. The feedback optimization module optimizes and adjusts the steady-state operation control strategy based on the information contained in the principal component eigenvector, combined with the system's operating objectives and constraints. For example, if the principal component eigenvector shows that the load distribution of the current power system has an unbalanced trend, and the operating status of some equipment is close to the critical value, the feedback optimization module will comprehensively consider this information and generate more reasonable control instructions, such as adjusting the power distribution in some areas, optimizing the operating parameters of the equipment, etc., to ensure the stable operation of the power system.

[0096] The system also builds an early warning module that includes an equipment health evaluation function. The equipment health evaluation function is constructed based on the equipment's historical operating data, current operating status, maintenance records and other information. By comprehensively considering these factors, the health status of the equipment can be accurately evaluated. The real-time control instructions are correlated with the equipment health. If the real-time control instructions may cause the equipment to operate beyond the range that its health can bear, or the equipment health drops to a certain level, the early warning module will output a graded early warning signal. For example, when the equipment health is lower than a certain threshold and the current control instructions may further increase the burden on the equipment, the early warning module will issue an early warning. Early warning signals are divided into different levels. For example, a first-level warning indicates that the equipment has serious risks and immediate measures need to be taken; a second-level warning indicates that the equipment has potential risks and needs to be closely monitored. The preset emergency response mechanism is triggered according to the graded early warning signal to generate equipment maintenance or load adjustment instructions. For example, when a first-level early warning signal is received, the system will immediately generate an equipment maintenance instruction and arrange professionals to conduct a comprehensive overhaul of the equipment; when a second-level early warning signal is received, part of the load may be transferred to other equipment by adjusting the load distribution to reduce the burden on the current equipment, thereby ensuring the stable operation of the power system.

[0097] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0098] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A full-dimensional real-time monitoring and stable operation linkage system for a power system, characterized in that, The system includes: A multi-dimensional data acquisition module, which is used to obtain an operation data set within a preset time range in the power system, and preprocess the operation data set according to a preset multi-dimensional data cleaning strategy to generate standardized operation data; A spatio-temporal feature mapping module, which is used to map the standardized operation data into a spatio-temporal feature space, and generate a spatio-temporal feature coordinate cluster based on a multi-dimensional spatio-temporal clustering algorithm. The spatio-temporal feature coordinate cluster includes multi-dimensional correlation relationships of equipment status, load distribution and environmental parameters; A dynamic monitoring module, which is used to adaptively adjust the threshold of the spatio-temporal feature coordinate cluster according to a preset anomaly detection rule, divide real-time monitoring intervals through a mixed integer linear programming algorithm, and extract an operation index set within each interval; A stable operation linkage control module, which is used to construct a linkage decision-making unit including a dynamic topology optimization network model, and iteratively update the parameters of the dynamic topology optimization network model by using the operation index set to generate a steady-state operation control strategy; A feedback optimization module, which is used to perform multi-source data fusion optimization on the steady-state operation control strategy according to a preset global sensitivity analysis model, and output a full-dimensional real-time regulation instruction for the power system; The adaptive threshold adjustment includes: Calculating the mean and variance of key operation indexes within each interval according to the historical operation data distribution of the real-time monitoring interval; Based on a dynamic sliding window algorithm, performing exponential smoothing correction on the mean and variance to generate a dynamic threshold benchmark; Using fuzzy logic rules to blur the interval boundaries of the dynamic threshold benchmark to generate an adaptive threshold interval; The real-time monitoring interval divided by the mixed integer linear programming algorithm includes: Defining the objective function as the weighted minimization of load fluctuation and equipment loss within the monitoring interval; Setting constraint conditions as grid topology connectivity, equipment capacity limit and environmental parameter tolerance range; Solving the objective function through the branch and bound method, and outputting an optimal monitoring interval division scheme; The global sensitivity analysis model includes: Constructing a multi-variable sensitivity calculation framework based on the Sobol index method to quantify the sensitivity of regulation instructions to each operation parameter; Generating a parameter perturbation sample set through Monte Carlo sampling, and calculating the variance contribution rate of each parameter; Selecting parameters with contribution rates higher than a preset threshold as core variables for optimized regulation.

2. The full-dimensional real-time monitoring and stable operation linkage system of the power system according to claim 1, characterized in that, Generating the spatio-temporal feature coordinate cluster based on the multi-dimensional spatio-temporal clustering algorithm includes: Extracting timestamps, equipment IDs and operation parameters in the standardized operation data to construct a multi-dimensional spatio-temporal matrix; Using a sliding window segmentation algorithm to segment the multi-dimensional spatio-temporal matrix to generate a set of time series sub-matrices; Adopting a density-based clustering model to perform spatial density analysis on the set of time series sub-matrices, and removing isolated noise points; Performing multi-level aggregation on the remaining sub-matrices through a hierarchical clustering algorithm to generate a feature coordinate cluster with spatio-temporal correlation.

3. The full-dimensional real-time monitoring and stable operation linkage system of the power system according to claim 1, wherein, The parameter iterative update of the dynamic topology optimization network model includes: Inputting the operation index set into the input layer of the dynamic topology optimization network model, and using a graph convolutional network to extract topological feature vectors; Weight is assigned to the topological feature vector through the attention mechanism to generate a device priority sequence; The backpropagation algorithm is combined with the genetic algorithm to optimize the network weights and update the decision parameters of the linkage decision unit.

4. The full-dimensional real-time monitoring and stable operation linkage system of the power system according to claim 1, characterized in that, The multi-dimensional data cleaning strategy includes: Identifying duplicate data segments in the operation data set and removing duplicates based on the timestamp alignment rule; Detecting missing data points and interpolating the missing data points using the spatio-temporal Kriging interpolation method; Normalizing the interpolated data to generate normalized operation data with a mean of zero and a variance of one.

5. The full-dimensional real-time monitoring and stable operation linkage system of the power system according to claim 1, characterized in that, The multi-source data fusion optimization includes: Integrating device sensor data, meteorological data, and user load prediction data to construct a heterogeneous data fusion matrix; Using the tensor decomposition algorithm to perform dimensionality reduction on the heterogeneous data fusion matrix and extract the principal component feature vectors; Inputting the principal component feature vectors into the feedback optimization module to generate optimized control instructions.

6. The full-dimensional real-time monitoring and stable operation linkage system of the power system according to any one of claims 1 to 5, characterized in that, The system further includes: Constructing an early warning module including a device health assessment function, associating the real-time control instructions with the device health, and outputting a graded early warning signal; Triggering a preset emergency response mechanism according to the graded early warning signal to generate device maintenance or load adjustment instructions.

7. An electronic device, characterized in that, Including: At least one processor; And A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the operations of the full-dimensional real-time monitoring and stable operation linkage system of the power system according to any one of claims 1 to 6.

Images