Mechanism data fusion power grid operation intelligent deduction and visual decision method and system

Abstract

The application belongs to the field of power grid safe operation, and discloses a mechanism data fusion power grid operation intelligent deduction and visual decision method and system, which comprises the following steps: obtaining power grid multi-source heterogeneous original data, performing preprocessing, and obtaining power grid operation characteristic data set; constructing a mechanism data fusion power grid topology dynamic analysis model, inputting the power grid operation characteristic data set for reasoning, and obtaining a real-time power grid topology matrix; extracting power grid safe operation core rules and converting them into a differentiable mathematical constraint form to obtain a safety constraint mathematical expression set; constructing a mechanism data fusion multi-scenario intelligent deduction model, embedding the safety constraint mathematical expression set into a model loss function as a regularization term, performing parallel calculation based on the real-time power grid topology matrix and the power grid operation characteristic data set, and obtaining a feasible operation mode scheme set; establishing a multi-objective comprehensive evaluation system to perform Pareto optimization sorting on the feasible operation mode scheme set, and outputting a global optimal operation scheme and a visual decision report.

Description

Technical Field

[0001] This invention belongs to the field of power grid safe operation technology, and in particular relates to a mechanism data fusion power grid operation intelligent deduction and visualization decision-making method and system. Background Technology

[0002] As the power system transitions to a new type of power system with a high proportion of new energy sources and high proportion of power electronic equipment, the strong randomness, volatility, and intermittency of new energy output lead to complex and ever-changing power grid operation modes, posing three core challenges to traditional dispatching technologies:

[0003] 1. Traditional topology analysis methods based on SCADA switch status suffer from acquisition delays and are susceptible to false alarms and missed alarms due to electromagnetic interference, equipment failures, and other factors. Under power grid fault conditions, the topology tracking response speed is severely insufficient.

[0004] 2. Traditional simulation methods based on power system simulation software such as PSASP and BPA rely on complex numerical solutions. A single N-1 safety check of the entire network takes more than 2 hours, which cannot meet the real-time scheduling requirements of frequent adjustments to the power grid operation mode under the high penetration rate of new energy. Although pure data-driven simulation methods can quickly generate schemes, the generated results do not meet the rules for safe operation of the power grid, and manual verification and iterative correction must be performed afterward. The overall efficiency still cannot meet the engineering requirements.

[0005] 3. Existing dispatching systems only output final operational plans in tabular and textual form, lacking multi-dimensional and intuitive decision-making visualization methods. They cannot clearly present the overall risk distribution of the power grid, changes in equipment margins, and the constraint satisfaction of plan adjustments. Dispatchers need to manually integrate and analyze multi-source heterogeneous data, which is not only time-consuming and labor-intensive but also prone to decision-making errors due to information overload. Especially in emergency fault handling scenarios, the speed and accuracy of decision response are difficult to guarantee.

[0006] Therefore, there is an urgent need for a mechanism-data fusion technology that can simultaneously achieve real-time and accurate analysis of power grid topology, efficient and safe simulation of operation modes, and intuitive visualization of the decision-making process, in order to solve the above-mentioned technical problems in the existing technology and meet the urgent needs of intelligent power grid dispatch under the new power system. Summary of the Invention

[0007] This application addresses the problems existing in the prior art by proposing a mechanism-data fusion method and system for intelligent simulation and visualization decision-making in power grid operation. It achieves real-time topology identification for physical compliance by embedding the physical mechanisms of the power grid as hard constraints into a data-driven topology analysis model. It quantifies safety rules using a power transfer distribution factor matrix and embeds it into the loss function of the simulation model, enabling compliance upon generating operational plans. A multi-dimensional visualization module intuitively displays decision-making plans, risk distribution, and constraint satisfaction. This shortens the simulation time for power grid operation modes, improves the accuracy of topology analysis, ensures comprehensive visualization of decision-making information, and enhances the intelligence and reliability of power grid dispatching.

[0008] To achieve the above objectives, this application provides the following technical solution:

[0009] Firstly, a mechanism-based data fusion method for intelligent inference and visualization decision-making in power grid operation includes the following steps: S1. Acquire multi-source heterogeneous raw data of the power grid and perform standardized preprocessing to obtain a standardized power grid operation feature dataset; S2. Construct a dynamic analytical model of the power grid topology that integrates the physical mechanism rules of the power grid and data-driven methods, input the power grid operation feature dataset for inference, and obtain a real-time power grid topology matrix that satisfies physical constraints; S3. Extract the core rules for safe operation of the power grid and transform them into differentiable mathematical constraint forms to obtain a set of differentiable mathematical expressions for safety constraints; S4. Construct a multi-scenario intelligent inference model that integrates power grid safety constraints and data-driven methods, use the set of mathematical expressions for safety constraints to constrain the model output, perform multi-scenario inference based on the real-time power grid topology matrix and the power grid operation feature dataset, and obtain a set of feasible operation mode schemes that satisfy safety constraints in all scenarios; S5. Establish a multi-objective comprehensive evaluation system to perform Pareto optimization ranking on the set of feasible operation mode schemes, output the globally optimal operation scheme, and generate a multi-dimensional visualized decision report that includes decision schemes, risk distribution, and constraint satisfaction.

[0010] Optionally, in step S1, the standardization preprocessing includes:

[0011] S11. Employ outlier detection and correction and wavelet threshold denoising algorithms to remove electromagnetic interference noise and outliers from the original data, and obtain the denoised effective dataset.

[0012] S12. Linear interpolation is used to complete short-term missing data, and WGAN-GP model is used to predict and complete long-term missing data to obtain a complete time series dataset.

[0013] S13. Establish a unified data metadata standard, perform format conversion and semantic alignment on data from different sources, and obtain a structured unified dataset;

[0014] S14. Extract key operating characteristics such as line load rate, node voltage deviation, and new energy penetration rate to obtain a standardized power grid operating characteristic dataset.

[0015] Optionally, step S2 includes:

[0016] S21. Abstract the power grid into a heterogeneous graph containing nodes and edges. The nodes include bus nodes, generator nodes, load nodes, and switch nodes. The edges include line edges, transformer edges, and switch connection edges. Assign node features containing node type, electrical parameters, and real-time operating status to each node. Assign edge features containing edge type, electrical parameters, and real-time operating status to each edge.

[0017] S22. Construct a topology analytical model architecture that includes a feature extraction layer, a mechanism constraint layer, and an output layer. The feature extraction layer uses a graph attention network to capture topological spatial features; the mechanism constraint layer embeds three types of physical constraints: topological connectivity, power balance, and equipment rated parameters; the output layer generates a power grid topology adjacency matrix.

[0018] S23. Design a composite loss function that integrates multi-dimensional physical constraints. The composite loss function is composed of a weighted sum of a prediction loss term and at least three types of physical constraint loss terms. The physical constraint loss terms include topology connectivity constraint loss, node power balance constraint loss, and equipment rated parameter constraint loss.

[0019] S24. Train the topology analytical model using historical power grid operation data. After training, input the power grid operation feature dataset and output a physically compliant real-time power grid topology matrix.

[0020] Optionally, the topology connectivity constraint loss is used to penalize prediction results that violate the basic connection rules of the power grid topology, including a bidirectional connection conflict penalty term and a switch state binarization penalty term;

[0021] The node power balance constraint loss is used to penalize prediction results that violate the node power conservation law, and is calculated based on the difference between the input power, output power, generator output, and load power of each node.

[0022] The equipment rated parameter constraint loss is used to penalize predictions that exceed the equipment's rated capacity, and is calculated based on the difference between the line current and the rated current, and the difference between the node voltage and the allowable voltage range.

[0023] Optionally, in step S23, the calculation formula for the composite loss function is as follows:

[0024]

[0025] in, This represents the total loss of the topological analytical model. Predicting loss for binary cross-entropy; Loss is due to topological connectivity constraints; The loss is due to node power balance constraints. Losses are due to constraints imposed by the equipment's rated parameters. , , These are the weighting coefficients.

[0026] Optionally, step S3 includes:

[0027] S31. Employ a fast calculation method based on the power transfer distribution factor matrix to quantify the N-1 fault safety constraints and construct the corresponding N-1 constraint loss function.

[0028] S32. Based on the relationship between node self-impedance and node voltage, quantify the short-circuit current constraint and construct the short-circuit current constraint loss function.

[0029] S33. Quantify voltage stability constraints based on voltage stability indices and construct the corresponding voltage stability constraint loss function;

[0030] S34. Summarize the above three types of constraint loss functions to obtain a set of differentiable mathematical expressions for safety constraints.

[0031] Optionally, step S31 includes:

[0032] S311. Calculate the power transfer distribution factor matrix of the power grid in advance;

[0033] S312. For any fault scenario where a line is disconnected, calculate the change in active power of the remaining lines using the power transfer distribution factor matrix.

[0034] S313. Based on the over-limit situation of line current and node voltage after the fault, construct an N-1 constrained loss function.

[0035] Optionally, step S4 includes:

[0036] S41. Using sampling methods, generate multiple typical operating scenarios covering fluctuations in new energy output, load fluctuations, and changes in meteorological conditions to obtain a set of typical operating scenarios;

[0037] S42. Construct a Transformer-GNN spatiotemporal fusion intelligent inference model, and embed the set of differentiable safety constraint mathematical expressions as regularization terms into the model's total loss function.

[0038] S43. Using the Spark distributed parallel computing framework, each typical scenario is assigned to different computing nodes, and the data-driven simulation model is called to perform independent simulations.

[0039] S44. Filter the deduction results that violate any security constraints, summarize all compliance results, and obtain a set of feasible operation mode solutions for physical compliance in all scenarios.

[0040] Optionally, in step S5, the multi-dimensional visualization decision report includes:

[0041] S541. Generate a heat map of global power grid operation risk, and use different colors to mark the risk level and corresponding value of each area;

[0042] S542. Generate a visualization of the optimal operation scheme adjustment measures, and use a bar chart to show the values ​​of generator output, transformer tap position, capacitor switching before and after adjustment, as well as the corresponding constraint satisfaction.

[0043] S543. Generate a visual dashboard for equipment margin, using a pointer-type dashboard to display pre-determined parameters such as load rate and voltage margin of heavy-load lines, main transformers and central nodes;

[0044] S544 provides an interactive visual query interface, supporting the viewing of detailed operating parameters and historical curves of any node or line by clicking;

[0045] S545. Summarize the above visualization content and generate the visualization decision report.

[0046] In a second aspect, the present invention provides a mechanism data fusion-based intelligent deduction and visualization decision-making system for power grid operation, used to implement the mechanism data fusion-based intelligent deduction and visualization decision-making method for power grid operation as described in the first aspect, comprising:

[0047] The data preprocessing module is used to acquire multi-source heterogeneous raw data of the power grid and perform standardized preprocessing to output a standardized power grid operation characteristic dataset.

[0048] The mechanism data fusion topology parsing module is used to construct a dynamic analytical model of power grid topology by fusion of mechanism data. It embeds the physical mechanism rules of the power grid as hard constraints into the data-driven model, inputs the power grid operation feature dataset for inference, and outputs a physically compliant real-time power grid topology matrix.

[0049] The safety rule quantification module is used to extract the core rules for safe operation of the power grid and transform them into differentiable mathematical constraints, outputting a set of differentiable mathematical expressions for safety constraints.

[0050] The mechanism data fusion intelligent inference module is used to construct a multi-scenario intelligent inference model of mechanism data fusion. It embeds the set of mathematical expressions of safety constraints as regularization terms into the model loss function, performs parallel calculations based on the real-time power grid topology matrix and the power grid operation feature dataset, and outputs a set of feasible operation mode schemes with full-scenario physical compliance.

[0051] The multi-objective optimization and visualization decision-making module is used to establish a multi-objective comprehensive evaluation system to perform Pareto optimization ranking on the set of feasible operating schemes, and output the globally optimal operating scheme and a multi-dimensional visualization decision report.

[0052] The beneficial effects of this application are as follows:

[0053] (1) This application uses the physical mechanism of the power grid as a hard constraint to embed the graph deep learning model, and ensures the symmetry of the adjacency matrix through the bidirectional connection contradiction penalty term, thereby improving the accuracy of topology analysis.

[0054] (2) This application uses the PTDF matrix to quickly quantify the safety rules and embeds the model loss function, which shortens the time for power grid operation mode simulation and effectively improves the effectiveness of the scheme.

[0055] (3) This application uses multi-dimensional visualization through risk heat map, adjustment comparison map, equipment dashboard and other methods to shorten the decision-making time of dispatchers and improve decision-making efficiency. Attached Figure Description

[0056] Figure 1 This is a schematic diagram illustrating the application scenarios of existing power grid operation and dispatching systems.

[0057] Figure 2 This is a schematic diagram of the mechanism data fusion power grid operation intelligent inference and visualization decision-making method according to Embodiment 1 of this application;

[0058] Figure 3 This is a schematic diagram of the mechanism data fusion power grid operation intelligent simulation and visualization decision-making system of Embodiment 2 of this application. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely one preferred embodiment of this application and are only used to explain this application. They do not limit the scope of protection of this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0060] like Figure 1The diagram illustrates a scenario application of an existing power grid operation and dispatching system. The dispatch console employs a multi-screen layout. The left side of the main display shows real-time operational data collected by the SCADA system in tabular form, while the right side displays a static power grid single-line diagram. A progress bar below indicates the N-1 safety verification process for the entire network, with a single verification taking nearly two hours. The two secondary screens below display operational mode schemes and historical data curves from a single dimension, respectively. In this scenario, the data is displayed in a fragmented manner, lacking an intuitive visualization of global risk distribution and equipment margins. Dispatchers must manually integrate information from multiple sources for decision-making, resulting in slow response times and a high risk of errors. This approach cannot meet the real-time dispatching needs of frequent adjustments to power grid operation modes under new power systems.

[0061] Example 1:

[0062] like Figure 2 As shown, a method for intelligent simulation and visualization decision-making of power grid operation based on mechanistic data fusion includes the following steps:

[0063] S1. Obtain the original data of the power grid from multiple sources and perform standardized preprocessing to obtain a standardized power grid operation characteristic dataset;

[0064] S2. Construct a dynamic analytical model of power grid topology that integrates the physical mechanism rules of the power grid and data-driven approaches. Input the power grid operation feature dataset for reasoning to obtain a real-time power grid topology matrix that satisfies physical constraints.

[0065] S3. Extract the core rules for safe operation of the power grid and transform them into differentiable mathematical constraint forms to obtain a set of differentiable mathematical expressions for safety constraints;

[0066] S4. Construct a multi-scenario intelligent simulation model that integrates power grid security constraints and data-driven approaches. Use the set of mathematical expressions for security constraints to constrain the model output. Based on the real-time power grid topology matrix and power grid operation feature dataset, perform multi-scenario simulations to obtain a set of feasible operation schemes that satisfy security constraints in all scenarios.

[0067] S5. Establish a multi-objective comprehensive evaluation system to perform Pareto optimization ranking on the set of feasible operation schemes, output the globally optimal operation scheme, and generate a multi-dimensional visualized decision report that includes decision schemes, risk distribution, and constraint satisfaction.

[0068] In step S1, the multi-source heterogeneous raw data of the power grid includes power grid operation monitoring data, equipment ledger data, new energy output forecast data, load forecast data, meteorological environment data, and historical fault data.

[0069] In step S1, the standardization preprocessing includes:

[0070] S11. Outlier detection and correction, along with wavelet thresholding denoising algorithms, are used to remove electromagnetic interference noise and outliers from the original data, resulting in a denoised, effective dataset. Specifically, outlier detection and correction are performed first: using 3... The (Laida) criterion, combined with wavelet high-frequency detail coefficients, is used to jointly identify abrupt outliers in SCADA data. This involves first setting a sliding window of 10 sampling points on the original time series data, and then calculating the mean of the data within the window. and standard deviation The value exceeds [ -3 , +3 Data points within the specified interval are marked as candidate outliers; then, the original data undergoes a 1-level db4 wavelet decomposition to extract the first-level high-frequency detail coefficients, and the positions where the amplitude of the detail coefficient is greater than 3 times the standard deviation of the coefficient are marked as outliers; the intersection of the two judgment results is taken as the final confirmed outlier.

[0071] For the identified outliers, the linear interpolation results of the two normal data points before and after them are used to replace and correct them, thus completing the outlier preprocessing.

[0072] Based on this, the db4 wavelet function is used to perform a three-level decomposition of the SCADA data, and a soft thresholding function is used to remove high-frequency noise. The formula for the soft thresholding function is shown below:

[0073]

[0074] in, The denoised wavelet coefficients of the j-th layer and the k-th wavelet; For the k-th wavelet coefficient of the j-th layer; The threshold value is used.

[0075] threshold The calculation is performed using a general threshold:

[0076]

[0077] Where is the noise standard deviation of the SCADA data, calculated using the robust median estimation method; N is the length of the data to be processed.

[0078] S12. Linear interpolation is used to complete short-term missing data, and the WGAN-GP model is used to predict and complete long-term missing data, resulting in a complete time-series dataset. For long-term missing data with a missing time of 5 minutes or more, the WGAN-GP model is used for prediction and completion. The WGAN-GP model uses an LSTM structure for the generator and a fully connected network for the discriminator. Gradient penalty terms are added during training to improve the model's stability and generation quality. The WGAN-GP model consists of a generator G and a discriminator D, trained adversarially. The generator generates the completed data, and the discriminator distinguishes between real and generated data. The loss function of WGAN-GP includes adversarial loss and gradient penalty terms. Specifically, the formula for calculating the discriminator loss is as follows:

[0079]

[0080] in, This represents the total loss value of the discriminator. To distribute the generated data The expected value of the following; Sample data generated by the generator; The data distribution fitted by the generator; The output score of the discriminator for the generated sample x; To determine the distribution of real data The expected value is given by the following formula; x is the actual data sample. This represents the true data distribution. The output score of the discriminator for the real sample x; These are the weight coefficients of the gradient penalty term; This is the gradient penalty term.

[0081] The formula for calculating the generator loss is as follows:

[0082]

[0083] in, This represents the total loss value of the generator; The expected value for scoring the generated samples by the discriminator.

[0084] The formula for calculating the gradient penalty term is as follows:

[0085]

[0086] in, This represents the gradient penalty loss value. For interpolation samples; The distribution of the interpolated samples; The gradient of the discriminator output with respect to the input; 1 represents the Euclidean norm; 1 represents the target gradient norm. This is for squaring operations.

[0087] As an optional implementation, the LSTM generator adopts a 2-layer structure with 128 hidden units in each layer. The input is historical data from the previous 24 hours, and the output is the completed data for the next 6 hours.

[0088] S13. Establish a unified data metadata standard, perform format conversion and semantic alignment on data from different sources to obtain a structured unified dataset; establish a unified data metadata standard, define metadata specifications for power grid equipment, operating parameters, meteorological data, etc., and convert data from different sources and in different formats into JSON-LD structured format to facilitate subsequent graph construction and model training.

[0089] S14. Extract key operational features such as line load factor, node voltage deviation, and renewable energy penetration rate to obtain a standardized power grid operation feature dataset. Key operational features are extracted from the preprocessed data, including line load factor, node voltage deviation, generator output fluctuation rate, renewable energy penetration rate, temperature, humidity, and wind speed.

[0090] Specifically, step S2 includes:

[0091] S21. Abstract the power grid into a heterogeneous graph containing nodes and edges. The nodes include bus nodes, generator nodes, load nodes, and switch nodes. The edges include line edges, transformer edges, and switch connection edges. Assign node features containing node type, electrical parameters, and real-time operating status to each node. Assign edge features containing edge type, electrical parameters, and real-time operating status to each edge.

[0092] S22. Construct a topology analytical model architecture that includes a feature extraction layer, a mechanism constraint layer, and an output layer. The feature extraction layer uses a graph attention network to capture topological spatial features; the mechanism constraint layer embeds three types of physical constraints: topological connectivity, power balance, and equipment rated parameters; the output layer generates a power grid topology adjacency matrix.

[0093] S23. Design a composite loss function that integrates multi-dimensional physical constraints. The composite loss function is composed of a weighted sum of a prediction loss term and at least three types of physical constraint loss terms. The physical constraint loss terms include topology connectivity constraint loss, node power balance constraint loss, and equipment rated parameter constraint loss.

[0094] S24. Train the topology analytical model using historical power grid operation data. After training, input the power grid operation feature dataset and output a physically compliant real-time power grid topology matrix.

[0095] In step S23, the topology connectivity constraint loss is used to penalize prediction results that violate the basic connection rules of the power grid topology, including a bidirectional connection conflict penalty term and a switch state binarization penalty term; the node power balance constraint loss is used to penalize prediction results that violate the node power conservation law, calculated based on the difference between the input power, output power, generator output, and load power of each node; the equipment rated parameter constraint loss is used to penalize prediction results that exceed the rated capacity of the equipment, calculated based on the difference between the line current and the rated current, and the difference between the node voltage and the allowable voltage range. Specifically, the calculation formula for the composite loss function is as follows:

[0096]

[0097] in, This represents the total loss of the topological analytical model. , where is the binary cross-entropy prediction loss, used to measure the deviation between the model's predicted switching state and the actual state; The loss is for topological connectivity constraints, including a bidirectional connection conflict penalty term and a switch state binarization penalty term. The node power balance constraint loss is calculated based on the difference between the node's input and output power. The loss is constrained by the equipment's rated parameters and is calculated based on the over-limit conditions of line current and node voltage. , , These are the weighting coefficients.

[0098] The formula for calculating the binary cross-entropy prediction loss is as follows:

[0099]

[0100] in, This represents the total number of switching sides in the power grid; Actual state of the switch; For predicting the state;

[0101] The formula for calculating the loss due to topological connectivity constraints is shown below:

[0102]

[0103] in, This represents the total number of nodes in the power grid. These are the adjacency matrix elements predicted by the model; This refers to the element in the j-th row and i-th column of the adjacency matrix predicted by the model. In this matrix, transmission lines in the power grid are bidirectional conducting elements, and the adjacency matrix describing the grid connection relationships must be a symmetric matrix. The first term is a bidirectional connection conflict penalty term, used to penalize the physical conflict result where the model predicts a unidirectional connection between node i and node j; switching devices in the power grid only have two definite binary states: open and closed, with no intermediate states. The second term is a switching state binarization penalty term, used to penalize fuzzy prediction results where the model outputs a switching state probability value close to 0.5, forcing the model output to be a definite binary state close to 0 (open) or 1 (closed), ensuring the determinism and engineering usability of the topology analysis results.

[0104] The formula for calculating the node power balance constraint loss is as follows:

[0105]

[0106] in, Let i be the set of neighbors of node i; Let be the active power flowing from node j to node i; Let be the active power flowing from node i to node j; The active power output of the generator connected to node i; The active load connected to node i.

[0107] The formula for calculating the constraint loss of the equipment's rated parameters is as follows:

[0108]

[0109] in, This represents the total number of lines in the power grid; This represents the actual current in line l. The rated current of line l; The actual voltage at node j; Let be the upper limit of the allowable voltage at node j; This represents the lower limit of the allowable voltage for node j.

[0110] In step S3, the core rules for safe operation of the power grid include N-1 fault safety constraints, short-circuit current limiting constraints, and voltage stability constraints; specifically, step S3 includes:

[0111] S31. Quantify the N-1 fault safety constraints using a fast calculation method based on the power transfer distribution factor matrix, and construct the corresponding N-1 constraint loss function; specifically, step S31 includes:

[0112] S311. Pre-calculate the power transfer distribution factor matrix of the power grid, its elements ,in, Let be the node-branch association vector of line l, with the vector dimension consistent with the total number of nodes in the power grid. Line l is defined as connecting the starting node i and the ending node j, and the positive reference direction is the flow of active power from the starting node to the ending node. Then the vector... The value is +1 at the position corresponding to the first node i, -1 at the position corresponding to the last node j, and 0 at the positions corresponding to the other nodes. Let be the node-branch association vector of line k; T is the matrix transpose symbol, and the vector dimension is the same as the total number of nodes in the power grid; it is agreed that line k connects the starting node m and the ending node n, and the positive reference direction is the flow of active power from the starting node to the ending node, then the vector The value is +1 at the position corresponding to the first node m, -1 at the position corresponding to the last node n, and 0 at the positions corresponding to the other nodes; B is the node susceptance matrix; For the change in active power of line k The influence coefficient on the active power of line l, the element sign calculated based on the above symbol convention, can directly reflect the power transfer direction. A positive value indicates that the power of line l after the fault is in the same positive direction as the reference, and a negative value indicates the opposite.

[0113] S312. For any fault scenario where line k is disconnected, calculate the change in active power of the remaining lines using the power transfer distribution factor matrix; when line k is disconnected, its active power changes from... The active power changes to 0, and the change in active power of the remaining line l is... ,in, The active power of line k before the fault;

[0114] The active power of line l after the fault is:

[0115]

[0116] in, This represents the active power of line l after line k is disconnected. This represents the active power of line l before the fault.

[0117] After the fault, the current in line l is:

[0118]

[0119] in, Let L be the current in line l after line k is disconnected; The line voltage amplitude of line l; Let be the power factor of line l.

[0120] S313. Based on the over-limit conditions of line current and node voltage after a fault, construct an N-1 constrained loss function, as shown in the following formula:

[0121]

[0122] in, The loss function is defined as an N-1 fault-safe constraint. This represents the total number of lines in the power grid; This represents the total number of nodes in the power grid. Let L be the current in line l after line k is disconnected; The rated current of line l; The voltage at node j after line k is disconnected; Let be the upper limit of the allowable voltage at node j; This represents the lower limit of the allowable voltage for node j.

[0123] S32. Based on the relationship between node self-impedance and node voltage, quantify the short-circuit current constraint and construct the short-circuit current constraint loss function; specifically, the three-phase short-circuit current of node j is:

[0124]

[0125] in, Let be the three-phase short-circuit current at node j; Let be the rated line voltage of node j; Let be the positive-sequence self-impedance of node j.

[0126] Short-circuit current constraint requirements:

[0127]

[0128] in, Let be the rated breaking current of the circuit breaker at node j.

[0129] The formula for calculating the short-circuit current constrained loss function is as follows:

[0130]

[0131] in, The loss function is defined by the short-circuit current limitation constraint. This represents the total number of nodes in the power grid.

[0132] S33. Quantify voltage stability constraints based on voltage stability indices and construct the corresponding voltage stability constraint loss function; specifically, the L index for a single load node j is:

[0133]

[0134] in, The voltage stability index L for load node j; For the set of generator nodes; Let J be the positive-sequence mutual impedance between nodes j and k. The active power output of generator k; Let be the voltage amplitude at load node j.

[0135] The overall voltage stability margin of the system is:

[0136]

[0137] in, For the overall voltage stability margin of the system; This is a set of load nodes. Requirements: ,in, This is the voltage stability margin threshold.

[0138] The formula for calculating the voltage stability constraint loss function is as follows:

[0139]

[0140] in, This is the voltage stability constraint loss function.

[0141] S34. Summarize the above three types of constraint loss functions to obtain a set of differentiable mathematical expressions for safety constraints.

[0142] Specifically, step S4 includes:

[0143] S41. The Latin hypercube sampling method is used to generate multiple typical operating scenarios covering the fluctuation of new energy output, load fluctuation and changes in meteorological conditions, and a set of typical operating scenarios is obtained.

[0144] S42. Construct a Transformer-GNN spatiotemporal fusion intelligent inference model, embedding the set of differentiable safety constraint mathematical expressions as regularization terms into the model's total loss function; the expression for the model's total loss function is:

[0145]

[0146] in, The total loss function for running the inference model; Task loss is used to measure the deviation between the predicted running state and the expected state. The loss is N-1 fault-safe constraint. The loss is limited by the short-circuit current constraint. Losses are due to voltage stability constraints. Economic losses are used to measure the total power generation cost and grid losses of the entire network. , , , These are the weighting coefficients.

[0147] The formula for calculating task loss is as follows:

[0148]

[0149] Where T is the prediction time step; The generator output is the predicted value at time t; Let t be the expected output of the generator. The node voltage prediction value at time t; Let be the expected value of the node voltage at time t.

[0150] The formula for calculating economic losses is as follows:

[0151]

[0152] in, This represents the total number of generator nodes. Let k be the unit power generation cost of generator k; Let be the active power output of generator k at time t; Total number of lines; Let be the resistance of line i; Let be the current in line i at time t; Unit conversion factor.

[0153] S43. Using the Spark distributed parallel computing framework, each typical scenario is assigned to different computing nodes, and the data-driven simulation model is called to perform independent simulations.

[0154] S44. Filter the deduction results that violate any security constraints, summarize all compliance results, and obtain a set of feasible operation mode solutions for physical compliance in all scenarios.

[0155] Specifically, step S5 includes:

[0156] S51. Establish a multi-objective comprehensive evaluation system that includes three dimensions: safety, economy, and reliability. Among them, the safety indicators include the maximum line load rate, the maximum node voltage deviation, the minimum short-circuit current margin, and the minimum voltage stability margin; the economic indicators include the total power generation cost of the entire network and the active power loss of the entire network; and the reliability indicators include the maximum load loss rate after N-1 fault and the average power supply availability rate.

[0157] S52. An improved NSGA-III algorithm is used to perform Pareto optimization sorting on the set of feasible operating schemes. An adaptive reference point adjustment mechanism is introduced to handle high-dimensional multi-objective optimization problems, obtaining the Pareto optimal solution set. Specifically, the improved NSGA-III algorithm introduces an adaptive reference point adjustment mechanism. Every 10 generations, the position and number of reference points are dynamically adjusted according to the current population distribution. Reference points not dominated by any individual are deleted, and new reference points are added in densely populated areas, improving the algorithm's search capability in high-dimensional multi-objective spaces. The adaptive reference point adjustment mechanism includes:

[0158] For generation t, the set of reference points is: Each reference point The set of related individuals is Define the reference point. Utilization rate:

[0159]

[0160] in, The utilization rate of the j-th reference point; The set of individuals associated with the j-th reference point; Population size. If < , If the threshold for determining an inefficient reference point is set, then the reference point is considered inefficient and marked for deletion.

[0161] Delete all that meet the criteria Reference points whose density is 0 and which have not been associated with any individual for two consecutive adjustment periods. Simultaneously, new reference points are added in densely populated areas. Densely populated areas are identified using the crowding distance variance. For any point p in the target space, its local density is estimated as:

[0162]

[0163] in, Let N be the local density of an individual p in its neighborhood; p is any individual in the population; N is the number of nearest neighbors. Let p be the set of the k nearest neighbors of individual p; Let p be the target vector of individual p.

[0164] In the region of highest density, a new reference point is generated using interpolation:

[0165]

[0166] in, For the newly generated reference point; , Two distinct individuals randomly selected from a dense region; The midpoint of the two target vectors; For small random perturbations; The directions are orthogonal.

[0167] To better match the scale of the target space with the reference point, each target dimension m is dynamically scaled:

[0168]

[0169] in, Let x be the value of individual x after scaling on the m-th target; Let x be the original function value of individual x on the m-th objective; , , representing the minimum and maximum values ​​of the current population at the m-th target, respectively; t is the current iteration number. The reference point is also transformed to the scaled space accordingly.

[0170]

[0171] in, The coordinates of the j-th reference point are scaled along the m-th dimension; Let be the coordinates of the original reference point in the m-th dimension.

[0172] When the number of reference points falls below a preset lower limit after deletion, new reference points are added using polar coordinates. A unit vector is generated on the scaled hyperplane.

[0173]

[0174] in, The components of a unit vector in dimension m; M is the total number of targets; For the k-th polar angle, a uniformly distributed reference direction is generated using a low-difference sequence:

[0175]

[0176] in, Let n be the nth value of the Halton sequence, and let k be the kth prime number as the cardinality; k is the target dimension index; and n is the sequence index.

[0177] The new reference point coordinates are:

[0178]

[0179] Where r is the magnitude of the new reference point; It is a median function; Let L2 norm be the scaled target vector of individual p; This is the current Pareto frontier set of individuals.

[0180] The final new reference point coordinates are:

[0181]

[0182] in, For an M-dimensional vector, the components are... Provided.

[0183] That is, the radius of the reference point is taken as the median of the magnitude of the target vector in the current Pareto front.

[0184] S53. Based on the Analytic Hierarchy Process (AHP), determine the weight vector of each indicator, calculate the comprehensive score of each scheme, and select the scheme with the highest comprehensive score as the global optimal operation scheme. The process of determining the indicator weights using the Analytic Hierarchy Process (AHP) is as follows: invite 10 experts with more than 5 years of experience in power grid dispatching to score the relative importance of 8 evaluation indicators on a scale of 1-9, and construct a judgment matrix; calculate the maximum eigenvalue and eigenvector of the judgment matrix, and perform a consistency test. When the consistency ratio CR < 0.1, normalize the eigenvector to obtain the final weight vector.

[0185] S54. Generate a multi-dimensional visual decision report. The visual decision report specifically includes:

[0186] S541. Generate a global power grid operation risk heat map, and use different colors to mark the risk level and corresponding value of each area; draw a global power grid operation risk heat map based on ECharts, using a continuous color scale in the range of 0-1, with green (0-0.3) representing low risk, yellow (0.3-0.7) representing medium risk, and red (0.7-1.0) representing high risk.

[0187] S542. Generate a visualization chart of the optimal operation plan adjustment measures, using a bar chart to show the values ​​of generator output, transformer tap position, and capacitor switching before and after adjustment, as well as the corresponding constraint satisfaction; use green to mark adjustments that satisfy the constraints and red to mark adjustments that are critical to the constraints;

[0188] S543. Generate a visual dashboard for equipment margin, using a pointer-type dashboard to display pre-determined parameters such as load rate and voltage margin of heavily loaded lines, main transformers, and central nodes; the pointer will automatically flash a warning when it exceeds the yellow zone; the thresholds of the visual dashboard for equipment margin are set as follows: green zone threshold 0%-70%, indicating sufficient equipment margin; yellow zone threshold 70%-90%, indicating that the equipment is close to its rated capacity; red zone threshold 90%-100%, indicating that the equipment is overloaded; the pointer will automatically flash a warning when it enters the yellow zone, and a red alarm pop-up window will appear when it enters the red zone.

[0189] S544 provides an interactive visual query interface, supporting the viewing of detailed operating parameters and historical curves of any node or line by clicking;

[0190] S545. Summarize the above visualization content and generate a PDF-formatted visualization decision report containing all charts and data.

[0191] Example 2:

[0192] like Figure 3 As shown, this embodiment provides a mechanism data fusion-based intelligent extrapolation and visualization decision-making system for power grid operation, used to implement the mechanism data fusion-based intelligent extrapolation and visualization decision-making method for power grid operation as described in Embodiment 1, including:

[0193] The data preprocessing module is used to acquire multi-source heterogeneous raw data of the power grid and perform standardized preprocessing to output a standardized power grid operation characteristic dataset.

[0194] The mechanism data fusion topology parsing module is used to construct a dynamic analytical model of power grid topology by fusion of mechanism data. It embeds the physical mechanism rules of the power grid as hard constraints into the data-driven model, inputs the power grid operation feature dataset for inference, and outputs a physically compliant real-time power grid topology matrix.

[0195] The safety rule quantification module is used to extract the core rules for safe operation of the power grid and transform them into differentiable mathematical constraints, outputting a set of differentiable mathematical expressions for safety constraints.

[0196] The mechanism data fusion intelligent inference module is used to construct a multi-scenario intelligent inference model of mechanism data fusion. It embeds the set of mathematical expressions of safety constraints as regularization terms into the model loss function, performs parallel calculations based on the real-time power grid topology matrix and the power grid operation feature dataset, and outputs a set of feasible operation mode schemes with full-scenario physical compliance.

[0197] The multi-objective optimization and visualization decision-making module is used to establish a multi-objective comprehensive evaluation system to perform Pareto optimization ranking on the set of feasible operating schemes, and output the globally optimal operating scheme and a multi-dimensional visualization decision report.

[0198] The above-described specific embodiments are preferred embodiments of the mechanism data fusion power grid operation intelligent deduction and visualization decision-making method and system of this application, and are not intended to limit the specific implementation scope of this application. The scope of this application includes but is not limited to the specific embodiments described above. All equivalent changes made in accordance with the shape and structure of this application are within the protection scope of this application.

Claims

1. A method for intelligent simulation and visual decision-making of power grid operation based on mechanistic data fusion, characterized in that, Includes the following steps: S1. Obtain the original data of the power grid from multiple sources and perform standardized preprocessing to obtain a standardized power grid operation characteristic dataset; S2. Construct a dynamic analytical model of power grid topology that integrates the physical mechanism rules of the power grid and data-driven approaches. Input the power grid operation feature dataset for reasoning to obtain a real-time power grid topology matrix that satisfies physical constraints. S3. Extract the core rules for safe operation of the power grid and transform them into differentiable mathematical constraint forms to obtain a set of differentiable mathematical expressions for safety constraints; S4. Construct a multi-scenario intelligent simulation model that integrates power grid security constraints and data-driven approaches. Use the set of mathematical expressions for security constraints to constrain the model output. Based on the real-time power grid topology matrix and the power grid operation feature dataset, perform multi-scenario simulations to obtain a set of feasible operation schemes that satisfy security constraints in all scenarios. S5. Establish a multi-objective comprehensive evaluation system to perform Pareto optimization ranking on the set of feasible operating schemes, output the globally optimal operating scheme, and generate a multi-dimensional visualized decision report that includes decision schemes, risk distribution, and constraint satisfaction.

2. The method for intelligent extrapolation and visual decision-making of power grid operation based on mechanistic data fusion according to claim 1, characterized in that, In step S1, the standardization preprocessing includes: S11. Employ outlier detection and correction and wavelet threshold denoising algorithms to remove electromagnetic interference noise and outliers from the original data, and obtain the denoised effective dataset. S12. Linear interpolation is used to complete short-term missing data, and WGAN-GP model is used to predict and complete long-term missing data to obtain a complete time series dataset. S13. Establish a unified data metadata standard, perform format conversion and semantic alignment on data from different sources, and obtain a structured unified dataset; S14. Extract key operating characteristics such as line load rate, node voltage deviation, and new energy penetration rate to obtain a standardized power grid operating characteristic dataset.

3. The method for intelligent extrapolation and visualization decision-making of power grid operation based on mechanistic data fusion according to claim 1, characterized in that, Step S2 includes: S21. Abstract the power grid into a heterogeneous graph containing nodes and edges. The nodes include bus nodes, generator nodes, load nodes, and switch nodes. The edges include line edges, transformer edges, and switch connection edges. Assign node features containing node type, electrical parameters, and real-time operating status to each node. Assign edge features containing edge type, electrical parameters, and real-time operating status to each edge. S22. Construct a topology analytical model architecture that includes a feature extraction layer, a mechanism constraint layer, and an output layer. The feature extraction layer uses a graph attention network to capture topological spatial features; the mechanism constraint layer embeds three types of physical constraints: topological connectivity, power balance, and equipment rated parameters; the output layer generates a power grid topology adjacency matrix. S23. Design a composite loss function that integrates multi-dimensional physical constraints. The composite loss function is composed of a weighted sum of a prediction loss term and at least three types of physical constraint loss terms. The physical constraint loss terms include topology connectivity constraint loss, node power balance constraint loss, and equipment rated parameter constraint loss. S24. Train the topology analytical model using historical power grid operation data. After training, input the power grid operation feature dataset and output a physically compliant real-time power grid topology matrix.

4. The method for intelligent extrapolation and visual decision-making of power grid operation based on mechanistic data fusion according to claim 3, characterized in that, The topology connectivity constraint loss is used to penalize prediction results that violate the basic connection rules of the power grid topology, including a bidirectional connection conflict penalty term and a switch state binarization penalty term; The node power balance constraint loss is used to penalize prediction results that violate the node power conservation law, and is calculated based on the difference between the input power, output power, generator output, and load power of each node. The equipment rated parameter constraint loss is used to penalize predictions that exceed the equipment's rated capacity, and is calculated based on the difference between the line current and the rated current, and the difference between the node voltage and the allowable voltage range.

5. The method for intelligent extrapolation and visual decision-making of power grid operation based on mechanistic data fusion according to claim 3, characterized in that, In step S23, the formula for calculating the composite loss function is as follows: in, This represents the total loss of the topological analytical model. Predicting loss for binary cross-entropy; Loss is due to topological connectivity constraints; The loss is due to node power balance constraints. Losses are due to constraints imposed by the equipment's rated parameters; , , These are the weighting coefficients.

6. The method for intelligent extrapolation and visual decision-making of power grid operation based on mechanistic data fusion according to claim 1, characterized in that, Step S3 includes: S31. Employ a fast calculation method based on the power transfer distribution factor matrix to quantify the N-1 fault safety constraints and construct the corresponding N-1 constraint loss function. S32. Based on the relationship between node self-impedance and node voltage, quantify the short-circuit current constraint and construct the short-circuit current constraint loss function. S33. Quantify voltage stability constraints based on voltage stability indices and construct the corresponding voltage stability constraint loss function; S34. Summarize the above three types of constraint loss functions to obtain a set of differentiable mathematical expressions for safety constraints.

7. The method for intelligent extrapolation and visual decision-making of power grid operation based on mechanistic data fusion according to claim 6, characterized in that, Step S31 includes: S311. Calculate the power transfer distribution factor matrix of the power grid in advance; S312. For any fault scenario where a line is disconnected, calculate the change in active power of the remaining lines using the power transfer distribution factor matrix. S313. Based on the over-limit situation of line current and node voltage after the fault, construct an N-1 constrained loss function.

8. The method for intelligent extrapolation and visualization decision-making of power grid operation based on mechanistic data fusion according to claim 1, characterized in that, Step S4 includes: S41. Using sampling methods, generate multiple typical operating scenarios covering fluctuations in new energy output, load fluctuations, and changes in meteorological conditions to obtain a set of typical operating scenarios; S42. Construct a Transformer-GNN spatiotemporal fusion intelligent inference model, and embed the set of differentiable safety constraint mathematical expressions as regularization terms into the model's total loss function. S43. Using the Spark distributed parallel computing framework, each typical scenario is assigned to different computing nodes, and the data-driven simulation model is called to perform independent simulations. S44. Filter the deduction results that violate any security constraints, summarize all compliance results, and obtain a set of feasible operation mode solutions for physical compliance in all scenarios.

9. The method for intelligent extrapolation and visualization decision-making of power grid operation based on mechanistic data fusion according to claim 1, characterized in that, In step S5, the multi-dimensional visualization decision report includes: S541. Generate a heat map of global power grid operation risk, and use different colors to mark the risk level and corresponding value of each area; S542. Generate a visualization of the optimal operation scheme adjustment measures, and use a bar chart to show the values ​​of generator output, transformer tap position, capacitor switching before and after adjustment, as well as the corresponding constraint satisfaction. S543. Generate a visual dashboard for equipment margin, using a pointer-type dashboard to display pre-determined parameters such as load rate and voltage margin of heavy-load lines, main transformers and central nodes; S544 provides an interactive visual query interface, supporting the viewing of detailed operating parameters and historical curves of any node or line by clicking; S545. Summarize the above visualization content and generate the visualization decision report.

10. A mechanism data fusion power grid operation intelligent simulation and visualization decision-making system, characterized in that, The method for implementing the mechanism data fusion power grid operation intelligent inference and visualization decision-making method as described in any one of claims 1-9 includes: The data preprocessing module is used to acquire multi-source heterogeneous raw data of the power grid and perform standardized preprocessing to output a standardized power grid operation characteristic dataset. The mechanism data fusion topology parsing module is used to construct a dynamic analytical model of power grid topology by fusion of mechanism data. It embeds the physical mechanism rules of the power grid as hard constraints into the data-driven model, inputs the power grid operation feature dataset for inference, and outputs a physically compliant real-time power grid topology matrix. The safety rule quantification module is used to extract the core rules for safe operation of the power grid and transform them into differentiable mathematical constraints, outputting a set of differentiable mathematical expressions for safety constraints. The mechanism data fusion intelligent inference module is used to construct a multi-scenario intelligent inference model of mechanism data fusion. It embeds the set of mathematical expressions of safety constraints as regularization terms into the model loss function, performs parallel calculations based on the real-time power grid topology matrix and the power grid operation feature dataset, and outputs a set of feasible operation mode schemes with full-scenario physical compliance. The multi-objective optimization and visualization decision-making module is used to establish a multi-objective comprehensive evaluation system to perform Pareto optimization ranking on the set of feasible operating schemes, and output the globally optimal operating scheme and a multi-dimensional visualization decision report.

Images