Three-phase imbalance phase-change switch switching control system based on graph neural network

Abstract

The invention discloses a three-phase imbalance phase-change switch switching control system based on a graph neural network, and relates to the technical field of phase-change switch control. In order to solve the problems that in the prior art, deep analysis of load dynamic association and time sequence laws is lacked, so that a control strategy lags behind, switching actions are frequent, the treatment effect is limited, instruction time delay is large, the packet loss rate is high, and accurate synchronization of a commutation moment is difficult to guarantee. According to the method, the accuracy and efficiency of load characteristic analysis are greatly improved by constructing the load association graph structure and identifying the dynamic association mode, the graph neural network is utilized to extract the space-time sequence characteristics of the power grid topology and the load change, and the multi-dimensional and refined optimization of the commutation strategy is realized in combination with the multi-target optimization model, so that the method is suitable for large-scale popularization and application. The Mesh self-organizing network based on the power TD-LTE private network ensures that the instruction can be synchronized in millisecond and accurately sent to the phase-change switch, and provides a solid communication guarantee for effective execution of the strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of phase-commutation switch control, and in particular to a three-phase unbalanced phase-commutation switch switching control system based on a graph neural network. Background Art

[0002] Traditional commutation switch control schemes are often based on static load data and fixed threshold strategies, lacking in-depth analysis of dynamic load correlations and timing patterns. This results in delayed control strategies, frequent switching, and limited control effectiveness. Patent application publication number CN117117910B discloses a method for automatically adjusting three-phase imbalance in a substation. While this patent can automatically adjust the three-phase imbalance in a substation based on the three-phase current in the substation and find the optimal commutation switch phase switching scheme, the following issues remain:

[0003] 1. Existing technologies rely solely on the instantaneous value and phase information of the substation bus current to calculate imbalance, without considering the temporal characteristics and spatial correlation of the load, and are unable to cope with complex changes in power consumption patterns.

[0004] 2. Failure to establish a multi-objective optimization system prevents comprehensive consideration of constraints such as the number of switch operations and the duration of phase sequence status maintenance. This can easily lead to frequent switch operations, shortening equipment life and increasing operation and maintenance costs.

[0005] 3. In actual distribution network environments, traditional single-path communication is susceptible to electromagnetic interference and network congestion, and there are risks of large command transmission delays and high packet loss rates. This makes it difficult to ensure accurate synchronization of commutation times, reducing the execution effect of control strategies. Summary of the Invention

[0006] The purpose of the present invention is to provide a three-phase unbalanced commutation switch control system based on graph neural network. Through precise load characteristic analysis, intelligent graph neural network decision-making and high-reliability real-time control execution, it can more effectively identify and alleviate three-phase imbalance problems, reduce losses and equipment risks caused by imbalance, and significantly improve the accuracy and optimization of commutation strategies to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A three-phase unbalanced commutation switch control system based on graph neural network includes:

[0009] The load characteristic analysis module is configured to obtain multi-dimensional load data of the substation, analyze the cycle and timing patterns of the busbar and branch loads in the substation, construct a load correlation graph structure, and determine the dynamic correlation pattern between single-phase loads and commutative loads;

[0010] A graph neural network decision module is configured to build a multi-layer graph neural network model based on the load association graph structure and, in combination with a multi-objective optimization model, generate an optimal commutation control strategy to control the commutation switches under three-phase unbalanced working conditions;

[0011] a real-time control execution module configured to convert the optimal commutation control strategy into a switch control instruction, determine the optimal phase switching moment, and establish a data transmission channel with the commutation switch to perform coordinated control of the commutation switch;

[0012] The performance evaluation feedback module is configured to collect the grid operation data before and after the commutation control of the commutation switch in real time, calculate the evaluation index to evaluate the collaborative control effect, and optimize the model parameters of the multi-objective optimization model based on the evaluation structure.

[0013] Furthermore, the load characteristics analysis module includes:

[0014] An electrical parameter acquisition unit is configured to acquire electrical parameter data of the substation bus and the phase-changing switch branch in real time, and obtain time sequence characteristic information corresponding to the electrical parameter data;

[0015] A load characteristic extraction unit is configured to determine the power consumption characteristic information of each load, identify each load in the electrical parameter data, extract the time series data of each load based on the time series characteristic matrix, and analyze the periodic characteristics and time series regularity of the load of the substation bus and the phase-changing switch branch;

[0016] A graph structure generating unit is configured to treat the bus nodes, each phase-changing switch branch node and the identified independent load nodes in the substation area as vertices in the graph structure, the connection relationship between the coupled vertices as the edge in the graph structure, and assign a weight to each edge to generate a load association graph structure;

[0017] The association pattern recognition unit is configured to recognize the dynamic association pattern between the single-phase load and the phase-switchable load at each branch node based on the load association graph structure.

[0018] Furthermore, the graph neural network decision module includes:

[0019] A graph neural network construction unit is configured to construct a multi-layer graph neural network model based on the load association graph structure, perform multi-layer information aggregation and feature conversion on the node features and edges between nodes in the load association graph structure, obtain the spatial-temporal characteristics of the power grid topology and load changes, and extract local and global power grid characteristics;

[0020] A multi-objective optimization modeling unit is configured to construct a multi-objective optimization model based on optimization objectives and constraints of three-phase imbalance, number of commutation switch operations, and phase sequence state maintenance time;

[0021] The strategy optimization decision unit is configured to input the acquired spatial-temporal characteristics into the multi-objective optimization model, and evaluate the optimal commutation strategy under the current power grid state based on the solution framework defined by the multi-objective optimization model with the goals of minimizing three-phase imbalance, constraining the number of switching operations, and maintaining the phase sequence state.

[0022] Furthermore, the load feature extraction unit specifically includes:

[0023] Principal component analysis is used to filter out redundancy and reduce the dimensionality of electrical parameter data, generating standardized data in a low-dimensional feature space, and constructing a time series feature matrix based on the standardized data;

[0024] Extract the time domain and frequency domain characteristics of electrical parameter data, combine the substation topology structure and distribution node attributes, and calculate the distribution topology weight value of each distribution node;

[0025] Perform weighted screening on the initial feature set based on the topological weight value, extract the key feature subset related to load operation, and construct a mapping relationship between the time series feature matrix and the key feature subset;

[0026] Based on the mapping relationship between the time series feature matrix and the key feature subset, the time series data of each load is extracted, and the load identification model is constructed in combination with the distribution topology weight value to obtain the power consumption feature information of each load;

[0027] Analyze the time evolution of power consumption characteristic information and identify the periodic characteristics and timing laws of the load.

[0028] Furthermore, the time evolution of power consumption characteristics is analyzed to identify the load cycle characteristics and timing patterns, including:

[0029] Perform time-scale normalization on the power consumption characteristic information of each load, uniformly adjust the power consumption characteristic information to a preset time interval, and form a standardized time series data set;

[0030] A multi-resolution analysis method is used to decompose the standardized time series data set to generate multiple decomposition levels. The energy value of each decomposition level is calculated to quantify the significance of the periodic characteristics of the corresponding time scale. Each decomposition level corresponds to a different time scale resolution.

[0031] Perform Fourier transform on the standardized time series data set to generate a frequency domain characteristic spectrum. Based on the frequency domain characteristic spectrum, the main frequency components and their amplitude distribution are determined to obtain the main period parameters of the load data.

[0032] Based on the feature extraction results of multi-resolution analysis and Fourier transform, a multi-dimensional periodic feature vector is constructed;

[0033] Compare the multi-dimensional periodic feature vector with the preset feature vector threshold to determine the load periodicity characteristics, including whether it has daily periodicity characteristics, weekly periodicity characteristics, or other periodicity characteristics;

[0034] Combining the mutation point information detected by wavelet transform with the time series change curve of power consumption characteristic parameters, the power mutation points and operating trend rules of the load in the preset time period are identified, and a load time series feature template reflecting the load operating rules is generated.

[0035] Furthermore, the real-time control execution module includes:

[0036] an instruction generation unit configured to parse the optimal commutation control strategy into a discretized switching action sequence and generate a control instruction set based on a phase sequence switching constraint of the power system;

[0037] The control transmission unit is configured to build a Mesh self-organizing network based on the power TD-LTE private network and ensure the synchronization of the command issuance time and the optimal phase switching time based on the time synchronization protocol;

[0038] The security protection unit is configured to monitor the communication link status and network traffic fluctuations of the Mesh self-organizing network in real time, detect abnormal behavior, and build an instruction integrity verification mechanism to perform integrity verification on the received instructions.

[0039] Furthermore, the control transmission unit further includes:

[0040] The link quality assessment subunit is configured to collect link parameters between Mesh network nodes in real time, perform link quality assessment based on the constructed link quality assessment model, and dynamically adjust the priority of the data transmission link based on the assessment results;

[0041] an adaptive retransmission control subunit, configured to establish a selective retransmission mechanism and match a corresponding retransmission strategy according to the priority of the data transmission channel;

[0042] The multi-path transmission optimization sub-unit is configured to generate multiple redundant transmission links based on the Mesh network topology structure, and allocate corresponding control instruction transmission traffic to each redundant transmission link based on the evaluation result of the link quality evaluation sub-unit.

[0043] Furthermore, the performance evaluation feedback module includes:

[0044] Link data acquisition unit, configured to collect real-time grid operation data before and after commutation control, switch action logs, and Mesh network communication data to construct a multi-dimensional data set;

[0045] a performance evaluation unit configured to calculate evaluation indicators based on the multidimensional data set and quantify the influence weight of each evaluation indicator on system performance;

[0046] The adaptive optimization feedback module is configured to generate a model optimization signal based on the performance evaluation results. When the three-phase imbalance improvement rate is lower than the expected threshold, the weight coefficient of the multi-objective optimization model is adjusted; when the switching action accuracy rate does not meet the standard, the instruction generation rule of the instruction generation unit is corrected.

[0047] Furthermore, the operation process of the performance evaluation unit includes:

[0048] Retrieving the grid operation data before and after each phase change control in the historical data, and obtaining the fluctuation amplitude ratio of the data transient fluctuation through the transient fluctuation of the grid operation data before and after the phase change control;

[0049] Retrieve Mesh network communication data from historical data;

[0050] Retrieving the communication delay at each communication moment from the Mesh network communication data;

[0051] Retrieving a communication time period in which a delay occurs, and using the communication time period in which a delay occurs to obtain a communication delay impact coefficient; wherein the communication delay impact coefficient is an average value of the ratio between the theoretical communication duration and the actual communication duration corresponding to all communication delays;

[0052] Retrieve the switch action log;

[0053] Determining whether there is a switch action fault from the switch action log;

[0054] When a switch action failure is present in the switch action log, a safety impact coefficient corresponding to the switch action failure is retrieved from the database; wherein, when the safety impact corresponding to the switch action failure is weak, the safety impact coefficient is set to 0.5; when the safety impact corresponding to the switch action failure is moderate, the safety impact coefficient is set to 1.2; and when the safety impact corresponding to the switch action failure is strong, the safety impact coefficient is set to 1.8;

[0055] Obtaining an evaluation index compensation coefficient based on the fluctuation amplitude ratio of the transient fluctuation of the data, the communication delay impact coefficient, and the safety impact coefficient corresponding to the switch action failure;

[0056] Retrieving the acquired evaluation indicators, wherein the evaluation indicators are obtained by calculating the three-phase imbalance improvement rate, the switching operation accuracy rate, and the system stability improvement rate through the multidimensional data set;

[0057] The obtained evaluation index is compensated and adjusted using the evaluation index compensation coefficient; wherein the compensation adjustment evaluation index is K*Q, and K represents the evaluation index compensation coefficient; Q represents the evaluation index.

[0058] Furthermore, the evaluation index compensation coefficient is obtained according to the fluctuation amplitude ratio of the transient fluctuation of the data, the communication delay impact coefficient, and the safety impact coefficient corresponding to the switch action failure, including:

[0059] Retrieve the fluctuation amplitude ratio of the data transient fluctuations before and after each phase change control, the communication delay impact coefficient, and the safety impact coefficient corresponding to the switch action failure;

[0060] The fluctuation amplitude ratio of the data transient fluctuation corresponding to each phase change control is processed with the theoretically allowed maximum fluctuation amplitude ratio to obtain the fluctuation amplitude ratio;

[0061] Comparing the fluctuation amplitude ratios before and after each commutation control with the fluctuation amplitude ratios corresponding to the transient fluctuations of the data before and after the current commutation control, respectively, screening out the fluctuation amplitude ratios whose fluctuation amplitude ratios exceed the fluctuation amplitude ratios corresponding to the transient fluctuations of the data before and after the current commutation control as observed fluctuation amplitude ratio data;

[0062] The evaluation index compensation coefficient is obtained by using the observed fluctuation amplitude ratio data in combination with the communication delay influence coefficient and the safety influence coefficient corresponding to the switch action failure.

[0063] The evaluation index compensation coefficient is obtained by the following formula:

[0064]

[0065] Among them, K represents the compensation coefficient of the evaluation index; n represents the number of observed fluctuation amplitude ratio data; F i Indicates the data value corresponding to the i-th observed fluctuation ratio data; F d Indicates the fluctuation amplitude ratio of the data transient fluctuation before and after the current commutation control; F p Indicates that the fluctuation amplitude ratio does not exceed the average fluctuation amplitude ratio of the corresponding data transient fluctuation before and after the current phase change control; A p represents the average safety impact coefficient corresponding to the switch action failure; Y represents the communication delay impact coefficient.

[0066] Compared with the prior art, the present invention has the following beneficial effects:

[0067] By constructing a load correlation graph structure, quantifying the coupling relationship between nodes, and identifying dynamic correlation patterns, the dynamic behavior of the substation load is analyzed, greatly improving the accuracy and efficiency of load characteristic analysis. Compared with traditional commutation strategies based on rules or simple models, the present invention uses graph neural networks to extract the spatial-temporal characteristics of grid topology and load changes, and combines them with multi-objective optimization models to achieve multi-dimensional and refined optimization of the commutation strategy. It is not only smarter and more responsive, but also takes into account the economy, stability and equipment life of system operation, and improves the robustness and practicality of decision-making. The Mesh self-organizing network based on the power TD-LTE private network significantly improves the reliability, anti-interference ability and real-time performance of control command transmission, ensuring that the commands can be synchronously and accurately delivered to the commutation switches at the millisecond level, providing solid communication guarantee for the effective execution of the strategy, and can more effectively identify and alleviate three-phase imbalance problems, reducing losses and equipment risks caused by imbalance. At the same time, by optimizing the switch action strategy and ensuring communication reliability, the equipment life is extended and the operation and maintenance costs are reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 This is a module diagram of the three-phase unbalanced phase-changing switch switching control system based on graph neural network of the present invention. DETAILED DESCRIPTION

[0069] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0070] The present invention will be further described below with reference to specific embodiments:

[0071] Example 1

[0072] like Figure 1 As shown, the three-phase unbalanced phase-changing switch control system based on the graph neural network of this embodiment includes:

[0073] The load characteristic analysis module is configured to obtain multi-dimensional load data of the substation, analyze the cycle and timing patterns of the busbar and branch loads in the substation, construct a load correlation graph structure, and determine the dynamic correlation pattern between single-phase loads and commutative loads;

[0074] The graph neural network decision module is configured to build a multi-layer graph neural network model based on the load association graph structure. Combined with the multi-objective optimization model, it generates the optimal commutation control strategy with the goal of minimizing the three-phase imbalance and constraining the number of switching operations, and controls the commutation switches under three-phase unbalanced conditions;

[0075] A real-time control execution module is configured to convert the optimal commutation control strategy into a switch control instruction, determine the optimal phase switching time, establish a data transmission channel with the commutation switch, build a Mesh self-organizing network based on the power TD-LTE private network, and coordinately control the commutation switch;

[0076] The performance evaluation feedback module is configured to collect real-time grid operation data such as voltage, current, and power before and after the commutation control of the commutation switch, calculate evaluation indicators such as the three-phase imbalance improvement rate, switch action accuracy, and system stability improvement to evaluate the collaborative control effect, and optimize the model parameters of the multi-objective optimization model based on the evaluation structure.

[0077] In this embodiment, the load characteristic analysis module is based on graph structure modeling, integrating principal component analysis and time series feature mining, and can capture dynamic load correlations more accurately than traditional methods. The graph neural network decision module uses multi-layer graph neural networks and multi-objective optimization models to reduce three-phase imbalance and, at the same time, reduce invalid switching actions. The real-time control execution module overcomes the defects of traditional single-path communication through Mesh self-organizing networks and multi-path transmission technology, taking into account both communication efficiency and network security, and achieving millisecond-level synchronization. The performance evaluation feedback module constructs a closed-loop optimization mechanism, dynamically adjusts model parameters and control rules, and effectively solves the existing three-phase imbalance management problem.

[0078] In this embodiment, the load characteristics analysis module includes:

[0079] An electrical parameter acquisition unit is configured to acquire electrical parameter data of the substation bus and the phase-changing switch branch in real time, and obtain time sequence characteristic information corresponding to the electrical parameter data;

[0080] A load characteristic extraction unit is configured to determine the power consumption characteristic information of each load, identify each load in the electrical parameter data, extract the time series data of each load based on the time series characteristic matrix, and analyze the periodic characteristics and time series regularity of the load of the substation bus and the phase-changing switch branch;

[0081] A graph structure generating unit is configured to treat the bus nodes, each phase-changing switch branch node and the identified independent load nodes in the substation area as vertices in the graph structure, the connection relationship between the coupled vertices as the edge in the graph structure, and assign a weight to each edge to generate a load association graph structure;

[0082] The association pattern recognition unit is configured to recognize the dynamic association pattern between the single-phase load and the phase-switchable load at each branch node based on the load association graph structure.

[0083] In this embodiment, through in-depth identification and analysis of the electricity consumption behavior of a single load, the dynamic behavior of various loads in the substation is accurately characterized, which can effectively improve the dynamic changes of loads in the substation, especially the accuracy and efficiency of identifying key loads that may cause or aggravate three-phase imbalance. At the same time, the physical grid topology and load characteristics are combined to construct a load association graph structure including busbars, branches and independent load nodes, and the coupling relationship between nodes is quantified by the weight of the edges. The association pattern between single-phase loads and switchable loads is dynamically identified, and the inherent connection of load changes is understood. The complexity of the global load information that needs to be processed is greatly reduced. On the premise of effectively improving the characterization strength of the influence of the graph node weight value on the load unit, the computational complexity of the subsequent graph neural network model is minimized, the utilization rate of computing resources is optimized, and the energy-saving efficiency of system operation is improved.

[0084] In this embodiment, the graph neural network decision module includes:

[0085] A graph neural network construction unit is configured to construct a multi-layer graph neural network model based on the load association graph structure, perform multi-layer information aggregation and feature conversion on the node features and edges between nodes in the load association graph structure, obtain the spatial-temporal characteristics of the power grid topology and load changes, and extract local and global power grid characteristics;

[0086] A multi-objective optimization modeling unit is configured to construct a multi-objective optimization model based on optimization objectives and constraints of three-phase imbalance, number of commutation switch operations, and phase sequence state maintenance time;

[0087] A strategy optimization decision unit is configured to input the acquired spatial-temporal characteristics into a multi-objective optimization model and evaluate the optimal commutation strategy under the current grid state based on a solution framework defined by the multi-objective optimization model with the objectives of minimizing three-phase imbalance, constraining the number of switching operations, and maintaining the phase sequence state.

[0088] In this embodiment, by deeply mining the load association graph structure, not only the spatial-temporal characteristics of the grid topology and load changes are obtained, but also the local and global grid characteristics are extracted at the same time, realizing a multi-dimensional and refined characterization of the complex grid state. At the same time, by introducing multi-objective optimization modeling, key operating indicators such as minimization of three-phase imbalance, constraint on the number of switch actions, and phase sequence state maintenance time are incorporated into a unified decision-making framework, which can effectively improve the comprehensive optimization level of the commutation strategy, ensure that the decision results not only focus on alleviating the imbalance, but also take into account the switch life and system stability, thereby improving the robustness and practicality of the decision. The spatial-temporal features extracted by the graph neural network are directly used to solve the multi-objective optimization model, rather than using a separate, rule-based or experience-based method. This can effectively improve the mapping efficiency from features to strategies and the intelligence level of decision-making, enabling the system to respond more quickly and accurately to changes in the grid state, and output commutation instructions that are highly matched with actual operating requirements.

[0089] In this embodiment, the real-time control execution module includes:

[0090] an instruction generation unit configured to parse the optimal commutation control strategy into a discrete switching action sequence, and generate a control instruction set including a phase switching time, a switch number, and a target phase based on a phase sequence switching constraint condition of the power system;

[0091] The control transmission unit is configured to build a Mesh self-organizing network based on the power TD-LTE private network, reliably transmit the generated control instruction set, and ensure microsecond-level synchronization between the instruction issuance time and the optimal phase switching time based on the time synchronization protocol (IEEE1588);

[0092] The security protection unit is configured to monitor the communication link status and network traffic fluctuations of the Mesh self-organizing network in real time, detect abnormal behavior, and build an instruction integrity verification mechanism to perform integrity verification on received instructions;

[0093] In this embodiment, the performance evaluation feedback module includes:

[0094] The link data acquisition unit is configured to collect real-time grid operation data (voltage, current, power), switch action logs, and Mesh network communication data before and after commutation control, and construct a multi-dimensional dataset containing timestamps, node IDs, and data types;

[0095] a performance evaluation unit configured to calculate evaluation indicators such as a three-phase imbalance improvement rate, a switch action accuracy rate (a ratio of consistency between actual action and instruction), and a system stability improvement rate (based on a change in a standard deviation of voltage fluctuation) based on the multidimensional data set, and quantify the influence weight of each evaluation indicator on system performance through grey correlation analysis;

[0096] The adaptive optimization feedback module is configured to generate a model optimization signal based on the performance evaluation results. When the three-phase imbalance improvement rate is lower than the expected threshold, the weight coefficient of the multi-objective optimization model is adjusted; when the switching action accuracy does not meet the standard, the instruction generation rules of the instruction generation unit are corrected; and the parameters of the multi-layer graph neural network model are iteratively optimized to form an "evaluation-feedback-optimization" closed-loop control mechanism.

[0097] In this embodiment, the real-time control execution module ensures that the optimal commutation strategy can be executed accurately, timely and safely through instruction generation, high-reliability and high-synchronization instruction transmission, and communication security protection; the performance evaluation feedback module realizes continuous monitoring, analysis and improvement of the system operation effect through refined performance quantification evaluation and closed-loop adaptive optimization mechanism, so that the entire control system can continuously adapt to changes in the power grid and continuously optimize control performance, thereby more effectively solving the three-phase imbalance problem and ensuring the stable and efficient operation of the power system.

[0098] Specifically, the operation process of the performance evaluation unit includes:

[0099] Retrieving the grid operation data before and after each phase change control in the historical data, and obtaining the fluctuation amplitude ratio of the data transient fluctuation through the transient fluctuation of the grid operation data before and after the phase change control;

[0100] Retrieve Mesh network communication data from historical data;

[0101] Retrieving the communication delay at each communication moment from the Mesh network communication data;

[0102] Retrieving a communication time period in which a delay occurs, and using the communication time period in which a delay occurs to obtain a communication delay impact coefficient; wherein the communication delay impact coefficient is an average value of the ratio between the theoretical communication duration and the actual communication duration corresponding to all communication delays;

[0103] Retrieve the switch action log;

[0104] Determining whether there is a switch action fault from the switch action log;

[0105] When a switch action failure is present in the switch action log, a safety impact coefficient corresponding to the switch action failure is retrieved from the database; wherein, when the safety impact corresponding to the switch action failure is weak, the safety impact coefficient is set to 0.5; when the safety impact corresponding to the switch action failure is moderate, the safety impact coefficient is set to 1.2; and when the safety impact corresponding to the switch action failure is strong, the safety impact coefficient is set to 1.8;

[0106] Obtaining an evaluation index compensation coefficient based on the fluctuation amplitude ratio of the transient fluctuation of the data, the communication delay impact coefficient, and the safety impact coefficient corresponding to the switch action failure;

[0107] Retrieving the acquired evaluation indicators, wherein the evaluation indicators are obtained by calculating the three-phase imbalance improvement rate, the switching operation accuracy rate, and the system stability improvement rate through the multidimensional data set;

[0108] The obtained evaluation index is compensated and adjusted using the evaluation index compensation coefficient; wherein the compensation adjustment evaluation index is K*Q, and K represents the evaluation index compensation coefficient; Q represents the evaluation index.

[0109] The technical effects of the above-mentioned technical solution are as follows: Based on the transient fluctuation analysis of the grid operating data before and after commutation control, the stability changes of the grid during the control process can be quantitatively assessed. By calculating the fluctuation amplitude ratio, the transient response characteristics of the grid are accurately captured, avoiding misjudgment of the operating status due to the lack of data fluctuation assessment. Communication delay is extracted from the mesh network communication data. The impact of communication quality on the system is converted into a quantifiable indicator by averaging the ratio of theoretical to actual communication duration (i.e., the communication delay impact coefficient). This solves the problem of unmeasurable or ambiguous communication delay and provides data support for communication link optimization. Fault type is analyzed through switch operation logs, and differentiated safety impact coefficients (0.5, 1.2, and 1.8) are set according to the safety impact level (weak, moderate, and strong). This allows the impact of faults on the system to be quantified, changing the traditional "black-and-white" fault judgment model and improving the refinement of system safety risk assessment. The system integrates the grid fluctuation amplitude ratio, communication delay impact factor, and switch failure safety impact factor into evaluation metric compensation coefficients. This dynamically corrects basic evaluation metrics, such as the three-phase imbalance improvement rate and switch operation accuracy, to address the limitations of single-dimensional evaluation and ensure that the evaluation results are more aligned with the system's actual operating conditions. A dynamic compensation mechanism based on historical data and real-time fault information automatically adjusts evaluation metrics as the system's operating status changes. This addresses the lag in traditional evaluation models' response to dynamic factors such as transient fluctuations and communication anomalies, and improves the evaluation system's adaptability to complex operating conditions. Quantitative analysis of communication delays and switch failures enables early identification of potential risks impacting system performance (such as control lags caused by communication delays and safety hazards caused by switch failures). Correcting evaluation metrics with compensation coefficients provides more accurate decision-making for risk warning and prevention strategies, mitigating system operational risks. This comprehensive evaluation and compensation of multi-dimensional data comprehensively optimizes the system's performance evaluation logic from various perspectives, including grid control, communication quality, and equipment failures. This allows the system to adjust its operating strategy based on real-time evaluation results, indirectly improving three-phase imbalance improvement, switch operation accuracy, and system stability, ultimately enhancing operational reliability across the entire link.

[0110] Specifically, the evaluation index compensation coefficient is obtained according to the fluctuation amplitude ratio of the transient fluctuation of the data, the communication delay impact coefficient, and the safety impact coefficient corresponding to the switch action failure, including:

[0111] Retrieve the fluctuation amplitude ratio of the data transient fluctuations before and after each phase change control, the communication delay impact coefficient, and the safety impact coefficient corresponding to the switch action failure;

[0112] The fluctuation amplitude ratio of the data transient fluctuation corresponding to each phase change control is processed with the theoretically allowed maximum fluctuation amplitude ratio to obtain the fluctuation amplitude ratio;

[0113] Comparing the fluctuation amplitude ratios before and after each commutation control with the fluctuation amplitude ratios corresponding to the transient fluctuations of the data before and after the current commutation control, respectively, screening out the fluctuation amplitude ratios whose fluctuation amplitude ratios exceed the fluctuation amplitude ratios corresponding to the transient fluctuations of the data before and after the current commutation control as observed fluctuation amplitude ratio data;

[0114] The evaluation index compensation coefficient is obtained by using the observed fluctuation amplitude ratio data in combination with the communication delay influence coefficient and the safety influence coefficient corresponding to the switch action failure.

[0115] The evaluation index compensation coefficient is obtained by the following formula:

[0116]

[0117] Among them, K represents the compensation coefficient of the evaluation index; n represents the number of observed fluctuation amplitude ratio data; F i Indicates the data value corresponding to the i-th observed fluctuation ratio data; F d Indicates the fluctuation amplitude ratio of the data transient fluctuation before and after the current commutation control; F p Indicates that the fluctuation amplitude ratio does not exceed the average fluctuation amplitude ratio of the corresponding data transient fluctuation before and after the current phase change control; A p represents the average safety impact coefficient corresponding to the switch action failure; Y represents the communication delay impact coefficient. By focusing on the abnormal deviation degree of transient fluctuation of power grid. i -F d The "abnormal difference" of a single observed fluctuation relative to the current commutation condition is quantified. The exponential operation will amplify the impact of the difference (the larger the difference, the more significant the contribution). After summing and averaging, it reflects the comprehensive contribution of all abnormal fluctuation conditions to the steady-state deviation of the power grid, reflecting the "abnormal cumulative effect" of transient fluctuations in the power grid caused by commutation control. China A p is the averaged reflection of the safety impact of switch failure, eY The impact weight of the communication delay is magnified, and the square root of the product of the two reflects the degree of coordinated interference of "fault + communication anomaly". Used to measure the "benchmark deviation" of the current commutation condition, F d -F p It reflects the difference between the current commutation fluctuation and the historical "normal fluctuation level". The exponential operation strengthens the impact of this deviation. After taking the inverse, the more the current fluctuation deviates from the normal mean (the larger the difference), the smaller the result. In essence, it is a correction to the "steady-state reference" of the current commutation control. That is, if the current commutation fluctuation itself deviates from the normal level, its reliability as an evaluation benchmark is reduced, and its impact on the compensation result needs to be weakened through this coefficient.

[0118] The technical solution achieves the following: By comparing the transient fluctuation amplitude ratio of the data before and after each commutation control with the theoretically allowed maximum fluctuation amplitude ratio, the abstract degree of fluctuation is converted into a specific fluctuation amplitude ratio value, enabling a quantitative representation of the fluctuations in the grid's operating status. Compared to analyzing only raw data, this method more accurately captures subtle changes in grid operation and avoids ambiguity in fluctuation assessment. By comparing the fluctuation amplitude ratio before and after each commutation control with the previous one, data exceeding the current fluctuation amplitude ratio is selected as observed fluctuation amplitude ratio data, effectively identifying abnormal fluctuations in grid operation that deviate from normal conditions. This screening mechanism avoids the interference of large amounts of conventional data on key abnormal information, allowing the evaluation process to focus more on fluctuation data that may affect system performance, thereby improving the efficiency and accuracy of abnormal problem diagnosis. By combining the communication delay impact coefficient and the safety impact coefficient corresponding to switch operation failures, the observed fluctuation amplitude ratio data is comprehensively utilized to obtain the compensation coefficient for the evaluation indicator, overcoming the limitations of single-factor analysis. By organically integrating multiple factors such as grid fluctuations, communication quality, and equipment failures, and fully considering the correlation between these factors and their combined impact on system performance, the evaluation results better reflect the actual system's operating status. Evaluation indicator compensation coefficients are calculated based on real-time grid fluctuation data, communication status, and equipment failures. These coefficients automatically adjust the degree of correction based on dynamic changes in system operating conditions. When grid fluctuations intensify, communication delays increase, or switch failures occur, the compensation coefficients respond promptly, ensuring that evaluation indicators consistently meet actual operating requirements and enhancing the evaluation system's adaptability to complex and changing operating environments. Multi-dimensional data processing and a multi-factor comprehensive analysis mechanism effectively mitigate the errors inherent in traditional evaluation methods due to insufficient data utilization and incomplete consideration of factors. By screening key abnormal data and performing multi-factor collaborative calculations, the evaluation indicator compensation coefficients more accurately correct the original evaluation indicators, thereby improving the credibility of core evaluation indicators such as the three-phase imbalance improvement rate and switch operation accuracy, providing a more reliable basis for system performance optimization. Furthermore, focused analysis of abnormal fluctuation data, combined with communication and equipment failure factors, helps identify potential risks and hidden dangers in system operation.

[0119] Example 2

[0120] In order to better realize the refined analysis of the substation load and improve the accuracy of three-phase imbalance management, based on the above embodiments, the load feature extraction process is further optimized, and key technical links such as data dimensionality reduction, topology weighting and multi-scale analysis are added.

[0121] Specifically in this embodiment, the load feature extraction unit includes:

[0122] Principal component analysis is used to filter out redundancy and reduce the dimensionality of electrical parameter data, generating standardized data in a low-dimensional feature space, and constructing a time series feature matrix based on the standardized data;

[0123] Extract the time domain and frequency domain features of electrical parameter data, combine the substation topology structure and distribution node attributes, and calculate the distribution topology weight value of each distribution node. For example, key nodes close to the busbar are given higher weights.

[0124] The initial feature set is weighted and screened based on the topological weight value to extract a subset of key features related to load operation, such as three-phase current imbalance, active power fluctuation coefficient, and load mutation rate. A mapping relationship between the time series feature matrix and the key feature subset is constructed.

[0125] Based on the mapping relationship between the time series feature matrix and the key feature subset, the time series data of each load is extracted, and the load identification model is constructed in combination with the distribution topology weight value to obtain the power consumption characteristic information of each load, including the load type classification result and power consumption characteristic parameters;

[0126] Analyze the time evolution of power consumption characteristic information, identify the periodic characteristics (such as daily cycle, weekly cycle) and timing laws of the load (such as power mutation points during peak hours in the morning and evening).

[0127] In this embodiment, analyzing the temporal evolution of power consumption characteristic information and identifying the periodic characteristics and timing patterns of the load specifically includes:

[0128] Perform time-scale normalization on the power consumption characteristic information of each load, uniformly adjust the power consumption characteristic information to a preset time interval, and form a standardized time series data set;

[0129] Multi-resolution analysis methods, including wavelet transform, empirical mode decomposition or variational mode decomposition, are used to decompose the standardized time series data set, generate multiple decomposition levels, calculate the energy value of each decomposition level (the ratio of the sum of the squares of the decomposition coefficients at that level to the total number of decomposition coefficients), and quantify the significance of the periodic characteristics of the corresponding time scale. Each decomposition level corresponds to a different time scale resolution.

[0130] Perform Fourier transform on the standardized time series data set to generate a frequency domain characteristic spectrum. Based on the frequency domain characteristic spectrum, the main frequency components and their amplitude distribution are determined to obtain the main period parameters of the load data.

[0131] Based on the feature extraction results of multi-resolution analysis and Fourier transform, a multi-dimensional cycle feature vector is constructed, which includes the daily cycle intensity, weekly cycle intensity, main cycle components and cycle distribution entropy value.

[0132] Compare the multidimensional periodic feature vector with the preset feature vector threshold to determine the load periodicity characteristics, including whether it has daily periodic characteristics, weekly periodic characteristics, or other periodic characteristics. Specifically, compare the daily periodic intensity with the first preset threshold. If it exceeds, it is determined that the load has significant daily periodic characteristics; compare the weekly periodic intensity with the second preset threshold. If it exceeds, it is determined that the load has significant weekly periodic characteristics; based on the frequency value of the main periodic component, identify other periodic characteristics of the load, such as monthly cycle, seasonal cycle, etc.

[0133] Combining the mutation point information detected by wavelet transform with the time series change curve of power consumption characteristic parameters, the power mutation points and operating trend rules of the load in the preset time period (such as morning and evening peaks) are identified, and a load time series feature template reflecting the load operation rules is generated.

[0134] In this embodiment, principal component analysis is used to filter redundant electrical parameter data and reduce its dimensionality, thereby improving data processing efficiency and significantly reducing the amount of subsequent computation while retaining core features. The distribution topology weight is calculated based on the substation topology structure, and features are weighted and screened based on this weight. Compared with traditional uniform screening methods, this method improves the accuracy of key feature extraction and more accurately reflects the load operating status. A method combining multi-resolution analysis and Fourier transform is used to identify periodic features and effectively capture load variation patterns on different time scales such as daily, weekly, and monthly. By constructing a multidimensional periodic feature vector and comparing it with a preset threshold, the load periodicity characteristics can be quickly and accurately determined. Only a subset of key features and information from a portion of the decomposition level are used for analysis, greatly reducing computational complexity. While ensuring analysis accuracy, it also reduces computing resource utilization, significantly improving system operating efficiency and energy conservation. In addition, the generated load time series feature template provides a decision-making basis that is more in line with the actual operating status for the subsequent phase-changing control strategy formulation, effectively improving the accuracy and reliability of three-phase imbalance management.

[0135] Example 3

[0136] Traditional communication links often use single-path transmission, which is easily affected by the complex electromagnetic environment of the distribution network. There are problems such as large command delay and high packet loss rate, making it difficult to ensure accurate synchronization of phase switching moments.

[0137] To address the above problem, in this embodiment, the control transmission unit further includes:

[0138] The link quality assessment subunit is configured to collect link parameters such as signal strength, packet loss rate, and latency between Mesh network nodes in real time, perform link quality assessment based on a link quality assessment model constructed using a fuzzy comprehensive evaluation method, and dynamically adjust the priority of the data transmission link based on the assessment results;

[0139] An adaptive retransmission control subunit is configured to establish a selective retransmission mechanism, matching the corresponding retransmission strategy according to the priority of the data transmission channel. When the link quality is good, selective retransmission is used, and when the link quality deteriorates, it switches to fallback frame retransmission and avoids retransmission conflicts through an exponential backoff algorithm.

[0140] The multi-path transmission optimization sub-unit is configured to generate multiple redundant transmission links based on the Mesh network topology structure, and based on the evaluation results of the link quality evaluation sub-unit, adopt a load balancing algorithm (such as weighted round-robin WRR) to distribute the transmission traffic of control instructions among the redundant transmission links, thereby reducing the risk of congestion on a single path.

[0141] In this embodiment, by dynamically adjusting the priority of the data transmission channel, it is ensured that key information such as the phase-changing control instructions can be transmitted reliably and timely between the real-time control execution module and the phase-changing switch, and network-level protection is provided for the execution of the collaborative control strategy. Based on the selective retransmission mechanism, lost data packets are accurately retransmitted to ensure timely and reliable delivery of instructions, thereby ensuring that the phase-changing strategy issued by the real-time control execution module can accurately and correctly act on the target phase-changing switch. At the same time, redundant backup of instructions is achieved through multi-path collaborative transmission, further improving communication reliability.

[0142] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A three-phase unbalanced commutation switch control system based on graph neural network, characterized in that: include: The load characteristic analysis module is configured to obtain multi-dimensional load data of the substation, analyze the cycle and timing patterns of the busbar and branch loads in the substation, construct a load correlation graph structure, and determine the dynamic correlation pattern between single-phase loads and commutative loads; A graph neural network decision module is configured to build a multi-layer graph neural network model based on the load association graph structure and, in combination with a multi-objective optimization model, generate an optimal commutation control strategy to control the commutation switches under three-phase unbalanced working conditions; a real-time control execution module configured to convert the optimal commutation control strategy into a switch control instruction, determine the optimal phase switching moment, and establish a data transmission channel with the commutation switch to perform coordinated control of the commutation switch; The performance evaluation feedback module is configured to collect the grid operation data before and after the commutation control of the commutation switch in real time, calculate the evaluation index to evaluate the collaborative control effect, and optimize the model parameters of the multi-objective optimization model based on the evaluation structure.

2. The three-phase unbalanced commutation switch control system based on graph neural network according to claim 1, characterized in that: Load characteristics analysis module, including: An electrical parameter acquisition unit is configured to acquire electrical parameter data of the substation bus and the phase-changing switch branch in real time, and obtain time sequence characteristic information corresponding to the electrical parameter data; A load characteristic extraction unit is configured to determine the power consumption characteristic information of each load, identify each load in the electrical parameter data, extract the time series data of each load based on the time series characteristic matrix, and analyze the periodic characteristics and time series regularity of the load of the substation bus and the phase-changing switch branch; A graph structure generating unit is configured to treat the bus nodes, each phase-changing switch branch node and the identified independent load nodes in the substation area as vertices in the graph structure, the connection relationship between the coupled vertices as the edge in the graph structure, and assign a weight to each edge to generate a load association graph structure; The association pattern recognition unit is configured to recognize the dynamic association pattern between the single-phase load and the phase-switchable load at each branch node based on the load association graph structure.

3. The three-phase unbalanced commutation switch control system based on graph neural network according to claim 2, characterized in that: The graph neural network decision module includes: A graph neural network construction unit is configured to construct a multi-layer graph neural network model based on the load association graph structure, perform multi-layer information aggregation and feature conversion on the node features and edges between nodes in the load association graph structure, obtain the spatial-temporal characteristics of the power grid topology and load changes, and extract local and global power grid characteristics; A multi-objective optimization modeling unit is configured to construct a multi-objective optimization model based on optimization objectives and constraints of three-phase imbalance, number of commutation switch operations, and phase sequence state maintenance time; The strategy optimization decision unit is configured to input the acquired spatial-temporal characteristics into the multi-objective optimization model, and evaluate the optimal commutation strategy under the current power grid state based on the solution framework defined by the multi-objective optimization model with the goals of minimizing three-phase imbalance, constraining the number of switching operations, and maintaining the phase sequence state.

4. The three-phase unbalanced commutation switch control system based on graph neural network according to claim 2, characterized in that: The load feature extraction unit specifically includes: Principal component analysis is used to filter out redundancy and reduce the dimensionality of electrical parameter data, generating standardized data in a low-dimensional feature space, and constructing a time series feature matrix based on the standardized data; Extract the time domain and frequency domain characteristics of electrical parameter data, combine the substation topology structure and distribution node attributes, and calculate the distribution topology weight value of each distribution node; Perform weighted screening on the initial feature set based on the topological weight value, extract the key feature subset related to load operation, and construct a mapping relationship between the time series feature matrix and the key feature subset; Based on the mapping relationship between the time series feature matrix and the key feature subset, the time series data of each load is extracted, and the load identification model is constructed in combination with the distribution topology weight value to obtain the power consumption feature information of each load; Analyze the time evolution of power consumption characteristic information and identify the periodic characteristics and timing laws of the load.

5. The three-phase unbalanced phase-changing switch control system based on graph neural network according to claim 4, characterized in that: Analyze the time evolution of power consumption characteristics and identify the load cycle characteristics and timing patterns, including: Perform time-scale normalization on the power consumption characteristic information of each load, uniformly adjust the power consumption characteristic information to a preset time interval, and form a standardized time series data set; A multi-resolution analysis method is used to decompose the standardized time series data set to generate multiple decomposition levels. The energy value of each decomposition level is calculated to quantify the significance of the periodic characteristics of the corresponding time scale. Each decomposition level corresponds to a different time scale resolution. Perform Fourier transform on the standardized time series data set to generate a frequency domain characteristic spectrum. Based on the frequency domain characteristic spectrum, the main frequency components and their amplitude distribution are determined to obtain the main period parameters of the load data. Based on the feature extraction results of multi-resolution analysis and Fourier transform, a multi-dimensional periodic feature vector is constructed; Compare the multi-dimensional periodic feature vector with the preset feature vector threshold to determine the load periodicity characteristics of the load; Combining the mutation point information detected by wavelet transform with the time series change curve of power consumption characteristic parameters, the power mutation points and operating trend rules of the load in the preset time period are identified, and a load time series feature template reflecting the load operating rules is generated.

6. The three-phase unbalanced commutation switch control system based on graph neural network according to claim 1, characterized in that: Real-time control execution module, including: an instruction generation unit configured to parse the optimal commutation control strategy into a discretized switching action sequence and generate a control instruction set based on a phase sequence switching constraint of the power system; The control transmission unit is configured to build a Mesh self-organizing network based on the power TD-LTE private network and ensure the synchronization of the command issuance time and the optimal phase switching time based on the time synchronization protocol; The security protection unit is configured to monitor the communication link status and network traffic fluctuations of the Mesh self-organizing network in real time, detect abnormal behavior, and build an instruction integrity verification mechanism to perform integrity verification on the received instructions.

7. The three-phase unbalanced phase-changing switch control system based on graph neural network according to claim 6, characterized in that: The control transmission unit further includes: The link quality assessment subunit is configured to collect link parameters between Mesh network nodes in real time, perform link quality assessment based on the constructed link quality assessment model, and dynamically adjust the priority of the data transmission link based on the assessment results; an adaptive retransmission control subunit, configured to establish a selective retransmission mechanism and match a corresponding retransmission strategy according to the priority of the data transmission channel; The multi-path transmission optimization sub-unit is configured to generate multiple redundant transmission links based on the Mesh network topology structure, and allocate corresponding control instruction transmission traffic to each redundant transmission link based on the evaluation result of the link quality evaluation sub-unit.

8. The three-phase unbalanced commutation switch control system based on graph neural network according to claim 1, characterized in that: Performance evaluation feedback module, including: Link data acquisition unit, configured to collect real-time grid operation data before and after commutation control, switch action logs, and Mesh network communication data to construct a multi-dimensional data set; a performance evaluation unit configured to calculate evaluation indicators based on the multidimensional data set and quantify the influence weight of each evaluation indicator on system performance; The adaptive optimization feedback module is configured to generate a model optimization signal based on the performance evaluation results. When the three-phase imbalance improvement rate is lower than the expected threshold, the weight coefficient of the multi-objective optimization model is adjusted; when the switching action accuracy rate does not meet the standard, the instruction generation rule of the instruction generation unit is corrected.

9. The three-phase unbalanced commutation switch control system based on graph neural network according to claim 8, characterized in that: The operation process of the performance evaluation unit includes: Retrieving the grid operation data before and after each phase change control in the historical data, and obtaining the fluctuation amplitude ratio of the data transient fluctuation through the transient fluctuation of the grid operation data before and after the phase change control; Retrieve Mesh network communication data from historical data; Retrieving the communication delay at each communication moment from the Mesh network communication data; Retrieving a communication time period in which a delay occurs, and using the communication time period in which a delay occurs to obtain a communication delay impact coefficient; wherein the communication delay impact coefficient is an average value of the ratio between the theoretical communication duration and the actual communication duration corresponding to all communication delays; Retrieve the switch action log; Determining whether there is a switch action fault from the switch action log; When a switch action fault occurs in the switch action log, the safety impact coefficient corresponding to the switch action fault is retrieved from the database; Obtaining an evaluation index compensation coefficient based on the fluctuation amplitude ratio of the transient fluctuation of the data, the communication delay impact coefficient, and the safety impact coefficient corresponding to the switch action failure; Retrieving the acquired evaluation indicators, wherein the evaluation indicators are obtained by calculating the three-phase imbalance improvement rate, the switching operation accuracy rate, and the system stability improvement rate through the multidimensional data set; The obtained evaluation index is compensated and adjusted using the evaluation index compensation coefficient; wherein the compensation adjustment evaluation index is K*Q, and K represents the evaluation index compensation coefficient; Q represents the evaluation index.

10. The three-phase unbalanced commutation switch control system based on graph neural network according to claim 9, characterized in that: The evaluation index compensation coefficient is obtained according to the fluctuation amplitude ratio of the transient fluctuation of the data, the communication delay impact coefficient, and the safety impact coefficient corresponding to the switch action failure, including: Retrieve the fluctuation amplitude ratio of the data transient fluctuations before and after each phase change control, the communication delay impact coefficient, and the safety impact coefficient corresponding to the switch action failure; The fluctuation amplitude ratio of the data transient fluctuation corresponding to each phase change control is processed with the theoretically allowed maximum fluctuation amplitude ratio to obtain the fluctuation amplitude ratio; Comparing the fluctuation amplitude ratios before and after each commutation control with the fluctuation amplitude ratios corresponding to the transient fluctuations of the data before and after the current commutation control, respectively, screening out the fluctuation amplitude ratios whose fluctuation amplitude ratios exceed the fluctuation amplitude ratios corresponding to the transient fluctuations of the data before and after the current commutation control as observed fluctuation amplitude ratio data; The evaluation index compensation coefficient is obtained by using the observed fluctuation amplitude ratio data in combination with the communication delay influence coefficient and the safety influence coefficient corresponding to the switch action failure. The evaluation index compensation coefficient is obtained by the following formula: Among them, K represents the compensation coefficient of the evaluation index; n represents the number of observed fluctuation amplitude ratio data; F i Indicates the data value corresponding to the i-th observed fluctuation ratio data; F d Indicates the fluctuation amplitude ratio of the data transient fluctuation before and after the current commutation control; F p Indicates that the fluctuation amplitude ratio does not exceed the average fluctuation amplitude ratio of the corresponding data transient fluctuation before and after the current phase change control; A p represents the average safety impact coefficient corresponding to the switch action failure; Y represents the communication delay impact coefficient.

Images