Power grid data processing method, device and system, and storage medium

Abstract

The application relates to a power grid data processing method, device and system and a storage medium, and relates to the technical field of power grid data processing. The method comprises the following steps: acquiring multi-dimensional power grid operation parameters under multiple operation conditions; inputting the multi-dimensional power grid operation data into a power grid state recognition model, so that the power grid state recognition model first calls an attention mechanism, performs self-adaptive weighting processing on each dimension of the multi-dimensional power grid operation parameters, and obtains weighted feature representation; then, a neural network is called, and the system state index of the power system is determined according to the weighted feature representation; the power grid state recognition model is trained through a loss function comprising a physical constraint loss; and finally, according to the attention weight of each dimension of the power grid operation parameters, target operation parameters with an attention weight higher than a preset weight condition are determined from the multi-dimensional power grid operation parameters, and the attention weight is used to represent the influence degree of each power grid operation parameter on the power grid state when the power grid state recognition model is running.

Description

Technical Field

[0001] This invention relates to the field of power grid data processing technology, and in particular to a power grid data processing method, apparatus, system and storage medium. Background Technology

[0002] With the continuous expansion of power system scale and the increasing complexity of its operation, the number of various operating parameters in the power grid is constantly increasing, and the internal coupling relationships within the system are becoming increasingly tight. Relevant power grid operation analysis methods largely rely on power flow calculations and sensitivity analyses, assessing system operational safety by calculating and comparing indicators such as node voltage and line power flow. However, in practical applications, due to the multi-dimensional nature of operating parameters and the complex interrelationships, the degree of influence of each parameter on the system state varies significantly under different operating conditions. Simply relying on fixed analysis methods or empirical rules makes it difficult to accurately identify the key operating parameters that play a dominant role in system safety, easily leading to one-sided analysis results or inaccurate judgments of key factors, thus affecting the effectiveness of operational decisions. Furthermore, how to effectively integrate the physical mechanisms of the power system in data-driven analysis, avoid model results deviating from actual operating patterns, and simultaneously achieve accurate identification of key operating parameters remains a pressing problem in the field of power system operation analysis. Summary of the Invention

[0003] This invention provides a power grid data processing method, apparatus, system, and storage medium to at least solve the problem in related technologies of the difficulty in accurately identifying key operating parameters that play a dominant role in system security. The technical solution of this invention is as follows: According to a first aspect of the present invention, a power grid data processing method is provided. The method includes: acquiring multi-dimensional power grid operating parameters under various operating conditions; the multi-dimensional power grid operating parameters include node load parameters, power generation output parameters, and operating modes of each node in the power system; inputting the multi-dimensional power grid operating data into a power grid state identification model, so that the power grid state identification model first calls an attention mechanism to adaptively weight the power grid operating parameters of each dimension in the multi-dimensional power grid operating parameters to obtain a weighted feature representation; then calls a neural network to determine the system state index of the power system based on the weighted feature representation; wherein the weighted feature representation is a weighted result based on the power grid operating parameters of each dimension and the attention weights of the power grid operating parameters of each dimension; the attention weights are used to characterize the degree of influence of each power grid operating parameter on the power grid state; the power grid state identification model is trained based on a loss function; the loss function includes physical constraint loss caused by not satisfying physical constraints; and, based on the attention weights of the power grid operating parameters of each dimension, determining target operating parameters from the multi-dimensional power grid operating parameters whose attention weights are higher than a preset weight condition.

[0004] The above scheme constructs a mapping model between operating parameters and system state, and introduces physical constraints during the training process to accurately extract key influencing factors under different operating conditions, thereby improving the physical consistency and reliability of the key operating parameter identification results.

[0005] In one technical solution, the method further includes: acquiring multi-dimensional historical operating parameters of the power system under various operating conditions, using these parameters as input samples; acquiring historical system state indicators under each operating condition, using these indicators as output samples corresponding to the input samples; constructing an input feature vector based on the input samples, and constructing an output feature vector corresponding to the input feature vector based on the output samples; adjusting and training the model parameters of the initial identification model based on the input feature vector until the loss function reaches a preset loss condition, and determining the initial identification model corresponding to the model parameters when the loss function reaches the preset loss condition as the power grid state identification model; the loss function also includes the model prediction loss between the output prediction index representing the system state index of the initial identification model and the corresponding output feature vector.

[0006] The above approach achieves supervised training of the model by constructing input-output sample pairs, thereby improving the accuracy of model predictions.

[0007] In one technical solution, both historical system status indicators and system status indicators include at least the following indicators: voltage safety indicators and line load indicators; the input feature vector is represented by the following formula;

[0008]

[0009] Among them, the power system includes load nodes and One generator, , The first Under the first operating condition Active and reactive loads of each load node; , The first Under the first operating condition The active and reactive power outputs of the generator nodes. For the first The operating mode feature vector corresponding to each operating condition; , , The first Load level coefficient, renewable energy output ratio, and tie line power level coefficient under each operating condition; Voltage safety indicators are characterized by the following formula;

[0010] in, Indicates the first Voltage safety indicators under various operating conditions; Indicates the first Under the first operating condition Voltage amplitude at each node; , They represent the first The upper and lower limits of the allowable voltage for each node; The line load index is represented by the following formula:

[0011] in, Indicates the first Line load indicators under various operating conditions; Indicates the total number of lines in the system; Indicates the first The allowable transmission capacity of each line; Indicates the first Under the first operating condition The power flow value of the line.

[0012] The above scheme quantifies the input features and system state indicators through specific formulas, providing a precise data foundation for model training.

[0013] In one technical solution, historical system state indicators of the power system under multi-dimensional historical operating parameters for each operating condition are obtained, including: power flow calculation based on the node admittance matrix, node injected power, and network topology of the power system under each operating condition and corresponding multi-dimensional historical operating parameters to obtain the historical voltage amplitude and historical line power flow of each node under each operating condition and corresponding multi-dimensional historical operating parameters; determining the historical voltage safety index of each node under each operating condition and corresponding multi-dimensional historical operating parameters based on the historical voltage amplitude of each node under each operating condition and corresponding multi-dimensional historical operating parameters; and determining the proportion of historical line power flow to allowable transmission capacity of each node under each operating condition and corresponding multi-dimensional historical operating parameters as the historical line load index of each node under each operating condition and corresponding multi-dimensional historical operating parameters.

[0014] The above scheme obtains historical system state indicators through power flow calculation, ensuring the consistency between training data and the actual operating patterns of the power system.

[0015] In one technical solution, physical constraints include node power balance constraints (composed of the active and reactive power of each node), voltage operating range constraints, and line transmission capacity constraints; physical constraint losses include node power balance losses, node voltage over-limit penalty functions, and line overload penalty functions; node power balance losses are determined based on the imbalance between the active and reactive power of each node; the node voltage over-limit penalty function is used to penalize situations where the voltage in the model prediction output exceeds a preset safe range; the line overload penalty function is used to penalize situations where the line power flow in the model prediction output exceeds a preset transmission capacity; wherein, physical constraint losses are characterized by the following formula;

[0016]

[0017]

[0018] in, These represent the imbalance of active and reactive power at the node, respectively. This represents the penalty function for exceeding the node voltage limit. This represents the line overload penalty function. Indicates the first Under the first operating condition Voltage amplitude at each node; , They represent the first The upper and lower limits of the allowable voltage for each node; Node voltage; For line power flow; This refers to the line capacity.

[0019] The above scheme introduces physical constraint loss, enabling the model to learn data features while satisfying the basic operating laws of the power system, thereby improving the reliability and engineering applicability of the key operating parameter identification results.

[0020] In one technical solution, based on the attention weights of various dimensions of power grid operating parameters, target operating parameters with attention weights higher than preset weight conditions are determined from multi-dimensional power grid operating parameters. This includes: determining the characteristic contribution of each dimension of power grid operating parameters based on disturbance analysis; determining the comprehensive importance index of each dimension of power grid operating parameters based on the attention weights and characteristic contribution of each dimension of power grid operating parameters; ranking each dimension of power grid operating parameters in descending order according to the comprehensive importance index; and selecting the power grid operating parameters ranked at the top of a preset percentage threshold from the descendingly ranked multi-dimensional power grid operating parameters as target operating parameters. These target operating parameters are used to guide power grid operation monitoring and dispatch decisions.

[0021] The above scheme combines attention weights and feature contribution to improve the accuracy and interpretability of key operational parameter identification.

[0022] In one technical solution, based on a disturbance analysis method, the characteristic contribution of each dimension of power grid operating parameters is determined, including: applying a disturbance of a preset magnitude to each dimension of power grid operating parameters to obtain each dimension of power grid operating variables corresponding to each dimension of power grid operating parameters; for any one dimension of power grid operating parameters, modifying the corresponding one dimension of power grid operating variables among the multiple dimensions of power grid operating variables to obtain modified multidimensional power grid operating variables; inputting the modified multidimensional power grid operating variables into a power grid state identification model to obtain the system state index after disturbance of any one dimension of power grid operating parameters; and determining the characteristic contribution of each dimension of power grid operating parameters based on the output index difference between the system state index after disturbance of each dimension of power grid operating parameters and the system state index.

[0023] The above scheme quantifies the impact of each operating parameter on the system state through disturbance analysis, providing a quantitative basis for the identification of key parameters.

[0024] According to a second aspect of the present invention, a power grid data processing apparatus is provided, which is capable of executing the power grid data processing method of the first aspect and any possible technical solution thereof, comprising: an acquisition unit configured to acquire multi-dimensional power grid operating parameters under various operating conditions; the multi-dimensional power grid operating parameters include node load parameters, power generation output parameters, and operating modes of each node in the power system; an identification unit configured to input the multi-dimensional power grid operating data into a power grid state identification model, so that the power grid state identification model first calls an attention mechanism to adaptively weight the power grid operating parameters of each dimension in the multi-dimensional power grid operating parameters to obtain a weighted feature representation; then calls a neural network to determine the system state index of the power system based on the weighted feature representation; wherein the weighted feature representation is a weighted result based on the power grid operating parameters of each dimension and the attention weights of the power grid operating parameters of each dimension; the attention weights are used to characterize the degree of influence of each power grid operating parameter on the power grid state; the power grid state identification model is trained based on a loss function; the loss function includes physical constraint loss caused by not satisfying physical constraints; and, based on the attention weights of the power grid operating parameters of each dimension, a target operating parameter with an attention weight higher than a preset weight condition is determined from the multi-dimensional power grid operating parameters.

[0025] According to a third aspect of the present invention, a power grid data processing system is provided, wherein the power grid data processing system stores instructions that, when executed by a controller, enable the controller to execute a power grid data processing method as described in the first aspect and any possible technical solution thereof.

[0026] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which instructions are stored, such that when the instructions in the computer-readable storage medium are executed by a processor of a control device, the control device is able to perform a power grid data processing method as described in the first aspect and any possible technical solution thereof.

[0027] According to a fifth aspect of the embodiments of this application, a computer program product is provided, the computer program product including computer instructions, which, when executed on a control device, cause the control device to perform the power grid data processing method of the first aspect and any possible technical solution thereof.

[0028] The technical solutions provided by the embodiments of the present invention bring at least the following beneficial effects: by integrating deep neural networks and multi-head attention mechanisms, adaptive weighted analysis is performed on multi-dimensional operating parameters to achieve a fine characterization of the degree of influence of different operating parameters; at the same time, power flow physical constraints are introduced during the model training process, and the node power balance relationship and voltage and line operation boundary conditions are incorporated into the model optimization objective, so that the model takes into account the power system operation mechanism while performing data-driven learning, thereby ensuring the physical consistency and engineering reliability of the key operating parameter identification results.

[0029] Based on this, by combining attention weight and feature contribution analysis methods, the importance of operating parameters is ranked, which can dynamically identify key operating parameters that have a significant impact on the system state under different operating conditions, forming a complete analysis process covering data acquisition, model building, physical constraints, and key identification.

[0030] Compared with related technologies, this invention overcomes the limitations of traditional empirical rules or single-indicator analysis in accurately identifying key influencing factors in complex operating scenarios by introducing a combination of deep learning and attention mechanisms. This significantly improves the accuracy and adaptability of key operating parameter identification. Simultaneously, by incorporating power flow physical constraints, it effectively avoids the problem of results deviating from actual operating patterns that may occur with purely data-driven models, thus improving the credibility and engineering applicability of the analysis results. Furthermore, this invention can adapt to complex operating conditions such as changes in power grid operation modes, new energy access, and load fluctuations, possessing good generalization and scalability. It can provide strong support for power grid operation monitoring, risk assessment, and dispatching decisions, and has high application value.

[0031] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0032] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0033] Figure 1 This is a flowchart illustrating a power grid data processing method according to an exemplary embodiment; Figure 2 This is a block diagram illustrating a power grid data processing apparatus according to an exemplary embodiment; Figure 3 This is a schematic diagram of a control device according to an exemplary embodiment. Detailed Implementation

[0034] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0035] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0036] For ease of understanding, the power grid data processing method provided in this application will be described in detail below with reference to the accompanying drawings.

[0037] Figure 1 This is a flowchart illustrating a power grid data processing method according to an exemplary embodiment, such as... Figure 1 As shown, the power grid data processing method includes the following steps.

[0038] Step S100: Obtain multi-dimensional power grid operating parameters under various operating conditions.

[0039] Multidimensional power grid operating parameters include the node load parameters, power generation output parameters, and operating modes of each node in the power system.

[0040] Multidimensional power grid operating parameters are the basic data describing the operating status of the power system. They can be obtained in real time through Supervisory Control and Data Acquisition (SCADA) or Wide Area Measurement System (WAMS), or retrieved from historical databases.

[0041] The node load parameters specifically include the active and reactive loads of each load node, reflecting the electricity demand of that node.

[0042] The power generation output parameters specifically include the active and reactive power output of each generator node, reflecting the power supply capacity of the power source side.

[0043] The operating mode characterizes the topological state of the power grid, such as the connection and disconnection status of lines, the opening and closing status of switches, and the tap position of transformers. Different operating modes determine the power flow distribution path of the power grid.

[0044] It should be understood that the various operating conditions in this embodiment cover the operating status of the power grid at different time sections, such as peak load conditions, off-peak load conditions, new energy power generation conditions, and N-1 fault conditions. By acquiring diverse operating condition data, comprehensive data support can be provided for subsequent model training, ensuring the applicability of the model in different scenarios.

[0045] Step S200: Input multidimensional power grid operation data into the power grid state identification model so that the power grid state identification model first calls the attention mechanism to adaptively weight the power grid operation parameters of each dimension in the multidimensional power grid operation parameters to obtain weighted feature representations; then calls the neural network to determine the system state indicators of the power system based on the weighted feature representations.

[0046] The weighted feature is represented by the weighted result between the power grid operating parameters of each dimension and the attention weights of the power grid operating parameters of each dimension; the attention weights are used to characterize the degree of influence of each power grid operating parameter on the power grid state; the power grid state identification model is trained based on the loss function; the loss function includes the physical constraint loss caused by the failure to meet the physical constraints.

[0047] The power grid state identification model integrates an attention mechanism module and a neural network module. Logically, the model doesn't treat all input parameters equally; instead, it first filters and focuses on the multi-dimensional power grid operating parameters through an attention mechanism. This mechanism calculates the correlation between each dimension's parameter and the overall system state, outputting a set of attention weights. The physical meaning of these weights is crucial: a larger weight indicates a greater impact of that dimension's operating parameter on the current power grid state (e.g., voltage stability, line load conditions), meaning that the parameter is a key factor causing changes in the system state; conversely, a smaller weight indicates a relatively weaker impact. For example, in heavily loaded areas, the attention weights corresponding to the load parameters of nodes in these areas are typically significantly higher than those in lightly loaded areas.

[0048] Therefore, through this adaptive weighting process, the model can highlight key information and suppress redundant information, thus obtaining a weighted feature representation that better characterizes the key features of the system. Subsequently, the neural network module receives this weighted feature representation and, through multi-layer nonlinear mapping, outputs system state indicators of the power system, such as voltage safety indicators and line load indicators.

[0049] Furthermore, the power grid state identification model in this embodiment introduces physical constraint loss during the training process. As a physical entity, the operation of a power system must follow fundamental physical laws such as Kirchhoff's laws and Ohm's law. Traditional pure data-driven models (such as conventional neural networks) rely solely on data fitting, which may produce prediction results that violate physical laws (e.g., the predicted node voltage amplitude exceeds the physical feasible region, or the power imbalance is too large).

[0050] To address this issue, this embodiment adds a physical constraint loss term to the loss function. When the model's prediction results do not meet preset physical constraints (such as node power balance constraints, voltage operating range constraints, and line transmission capacity constraints), the physical constraint loss generates a penalty value, forcing the model to optimize in a direction that satisfies physical laws when updating parameters during backpropagation. This design not only improves the accuracy of model predictions but, more importantly, ensures the physical interpretability and engineering reliability of the prediction results, preventing model predictions from violating the fundamental laws of power systems.

[0051] Step S300: Based on the power grid state identification model, and according to the attention weights of the power grid operating parameters in each dimension, determine the target operating parameters whose attention weights are higher than the preset weight conditions from the multi-dimensional power grid operating parameters.

[0052] While the model outputs the system state indicators, the attention weights output by the attention mechanism module are also recorded. Since the attention weights characterize the degree of influence of parameters on the power grid state, target operating parameters can be screened by setting preset weight conditions (e.g., setting a weight threshold, or selecting the top N parameters by weight). These target operating parameters are the key parameters that have the most significant impact on the safe operation of the power grid under the current operating conditions. For example, if the reactive load parameter of a certain node has the highest attention weight and exceeds the preset threshold, it is identified as a target operating parameter, prompting dispatchers to pay close attention to the reactive power compensation situation of that node.

[0053] In this way, this embodiment enables the automatic identification of key influencing factors from massive operational data, providing precise guidance for power grid operation monitoring and dispatching decisions.

[0054] As one implementation method, the training process of the power grid state identification model is described in detail based on the above embodiments. The training of this model relies on high-quality sample data, and a complete chain of data preparation, label generation, and model training ensures that the model possesses accurate predictive capabilities.

[0055] The system acquires multidimensional historical operating parameters of the power system under various operating conditions, and uses these multidimensional historical operating parameters as input samples.

[0056] It should be understood that the sources of multidimensional historical operating parameters can be actual operating records stored in the power grid historical operating database, simulation data generated by a digital simulation system, or a combination of both. Multidimensional historical operating parameters cover the power grid operating status under different time segments, different load levels, and different start-up methods, thus constructing a comprehensive input sample set.

[0057] Subsequently, an input feature vector is constructed. Based on the input samples, an input feature vector is constructed. Specifically, for the k-th operating condition, the input feature vector not only includes direct operating parameters such as active and reactive loads, generator active and reactive outputs at each node, but also derived parameters that characterize the operating condition, such as load level coefficients, renewable energy output ratios, and tie-line power level coefficients. By combining the original operating parameters with the derived parameters, the operating state of the power grid can be described more comprehensively, improving the model's ability to extract features from complex operating conditions.

[0058] Simultaneously, historical system state indicators are acquired under multi-dimensional historical operating parameters for each operating condition of the power system, and these historical system state indicators are used as output samples corresponding to the input samples. This process is a crucial step in generating training labels.

[0059] This embodiment uses a power flow calculation method based on physical mechanisms to generate tags. Specifically, it includes: performing power flow calculations based on the node admittance matrix, node injected power, and network topology of the power system under each operating condition and corresponding multi-dimensional historical operating parameters, to obtain the historical voltage amplitude and historical line power flow of each node under each operating condition and corresponding multi-dimensional historical operating parameters.

[0060] Power flow calculation is the most fundamental calculation in power system analysis. Given a network structure, operating parameters, and load conditions, it solves for the steady-state operation of the power grid. By solving the nodal power balance equations, the voltage magnitude, phase angle, and power flow of each branch can be accurately calculated. Compared to directly using measurement data as labels, the label data generated based on power flow calculation has higher accuracy and physical consistency, effectively avoiding the interference of measurement errors on model training.

[0061] After obtaining the power flow calculation results, the historical voltage safety index of each node under each operating condition and the corresponding multi-dimensional historical operating parameters is further determined based on the historical voltage amplitude of each node under each operating condition and the corresponding multi-dimensional historical operating parameters.

[0062] Voltage safety indicators are used to quantify the degree to which node voltage deviates from the safe operating range. For example, this indicator can be constructed by calculating the deviation between the voltage amplitude and the upper and lower voltage limits. Simultaneously, the proportion of historical line power flow to allowable transmission capacity for each node under each operating condition and corresponding multi-dimensional historical operating parameters is determined as the historical line load indicator for each node under each operating condition and corresponding multi-dimensional historical operating parameters. The line load indicator reflects the load level of the lines and is another important indicator for assessing power grid security. Finally, the voltage safety indicator and the line load indicator are combined to form an output feature vector, which serves as a supervision signal for model training.

[0063] After constructing the input and output feature vectors, the model parameters of the initial identification model are adjusted and trained based on the input feature vectors. During training, the input feature vectors are input into the initial identification model, and the model outputs predicted system state indicators. The model prediction loss is constructed by calculating the difference between the predicted indicators and the output feature vectors. Simultaneously, as in the above embodiment, the loss function also includes physical constraint loss. The model uses the backpropagation algorithm to calculate the gradient using the total loss function, continuously updating the model's weights and bias parameters until the loss function reaches a preset loss condition. The initial identification model corresponding to the model parameters where the loss function reaches the preset loss condition is then determined as the power grid state identification model.

[0064] The preset loss condition can be that the value of the loss function is less than a certain preset threshold, or that the value of the loss function no longer decreases significantly in multiple consecutive training rounds. Through the above training process, the power grid state recognition model can not only learn the complex nonlinear mapping relationship between multidimensional power grid operating parameters and system state indicators, but also guide its prediction results to conform to the basic physical laws of the power system through the guidance of physical constraint loss, thereby ensuring the reliability and generalization ability of the model in practical applications.

[0065] As another implementation method, based on the above embodiments, the specific construction method of the mathematical model and physical constraint formulas involved in the power grid data processing is described in detail. Through specific mathematical expressions, the mapping relationship between power grid operating parameters and system state is quantified, and the physical laws of the power system are integrated into the model training process to ensure the physical interpretability of the data processing results.

[0066] Specifically, this embodiment uses the following formula to represent the construction of the input feature vector.

[0067]

[0068]

[0069] Among them, the power system includes Each load node The generator, for the first Operating conditions , The first Under the first operating condition Active and reactive loads of each load node; , The first Under the first operating condition The active and reactive power outputs of the generator nodes. For the first The operating mode feature vector corresponding to each operating condition; , , The first Load level coefficient, renewable energy output ratio, and tie line power level coefficient under each operating condition; Input feature vector under various operating conditions .

[0070] In the above implementation, the load level coefficient, the proportion of renewable energy output, and the tie-line power level coefficient are high-order abstractions of the original operating parameters. For example, the proportion of renewable energy output can intuitively reflect the impact of intermittent power sources on grid stability, while the tie-line power level coefficient characterizes the intensity of power exchange between regions. By introducing these features, the input feature vector not only includes the physical measurements of grid operation but also integrates key indicators characterizing operating conditions, thereby enriching the information dimensions of the model input and helping to improve the model's ability to perceive complex operating conditions.

[0071] In one implementation, the input samples can be derived from historical operating data, online monitoring data, offline simulation data, or any combination of the above. To ensure that the constructed dataset covers as many operating scenarios as possible, load levels, generator output, and operating modes can be perturbed or combined to form multiple sets of samples with different operating conditions.

[0072] As another implementation method, this embodiment mainly focuses on voltage safety indicators and line load indicators for system status indicators.

[0073] Specifically, for the first Under each operating condition, power flow calculations are performed based on the power system node admittance matrix, node injected power, and network topology to obtain the voltage at each node and the power flow on each line. The active and reactive power balance equations at each node can be expressed as follows.

[0074]

[0075]

[0076] in, , They represent the first The active and reactive power injected into each node; , They represent the first and the Voltage amplitude at each node; , These represent the conductance and susceptance components in the nodal admittance matrix, respectively. , They represent the first and the The voltage phase angle of each node.

[0077] After obtaining the power flow calculation results, indicators characterizing the system's operating status are extracted as output labels. Preferably, the system status indicators include at least voltage safety indicators and line load indicators.

[0078] The voltage safety index can be defined by the following formula.

[0079]

[0080] in, Indicates the first Voltage safety indicators under various operating conditions; Indicates the first Under the first operating condition Voltage amplitude at each node; , They represent the first The upper and lower limits of the allowed voltage for each node.

[0081] The formula The physical meaning is: when the node voltage is within the allowable range, The function takes a value of 0, meaning no penalty is applied; however, once the voltage exceeds the limit, whether it is above the upper limit or below the lower limit, the function will calculate the magnitude of the exceedance. By summing the exceedance magnitudes of all nodes, this indicator can quantify the degree to which the grid voltage deviates from the safe operating range under the current operating conditions; a larger value indicates a higher voltage safety risk.

[0082] The line load index can be defined by the following formula.

[0083]

[0084] in, Indicates the first Line load indicators under various operating conditions; Indicates the total number of lines in the system; Indicates the first The allowable transmission capacity of each line; Indicates the first Under the first operating condition The power flow value of this line. This indicator The most severe line load situation in the power grid is characterized by calculating the ratio of power flow to transmission capacity for each line and taking the maximum value. It should be understood that the line load index is not limited to taking the maximum value. In other implementations, the average or variance of the load rates of all lines can also be used to characterize the overall load level. The specific choice can be adjusted according to the actual focus.

[0085] In one implementation, the voltage safety index and the line load index can be combined to form an output label vector represented by the following formula. .

[0086]

[0087] Finally, the sample dataset is constructed:

[0088] in, A dataset representing the relationship between power system operating parameters and grid operating status; Indicates the first A set of input-output sample pairs corresponding to each operating condition.

[0089] Furthermore, a deep neural network and a multi-head attention mechanism are integrated to model the mapping relationship between operating parameters and system state, and adaptive weighting is applied to different operating parameters to characterize the degree of influence of each parameter. The specific process is as follows.

[0090] First, take the input feature vector As input to the model, features are mapped layer by layer through a multi-layer fully connected neural network. The calculation process can be expressed as the following formula.

[0091]

[0092]

[0093]

[0094] in, Indicates the first Hidden layer output; This represents the system state output predicted by the model; , They represent the first Layer weight matrix and bias vector; , These represent the weight matrix and bias vector of the output layer, respectively. Indicates the number of hidden layers; This represents a non-linear activation function.

[0095] The aforementioned deep neural network enables the modeling of the nonlinear mapping between operating parameters and system state.

[0096] Secondly, based on the deep neural network constructed above, a multi-head attention mechanism is introduced to adaptively weight the input operating parameters in order to characterize the degree of influence of different operating parameters on the system state.

[0097] Specifically, for the input feature vector We construct H independent feature weighted branches, each branch corresponding to a feature weight vector, and the calculation process is as follows.

[0098]

[0099] In the formula, This is the feature weight vector corresponding to the h-th weighted branch; Here, are the weight matrix and bias vector of the h-th branch, respectively; H is the number of weighted branches; This represents the normalization function, used to map weights to... Interval.

[0100] Based on the aforementioned weight vector, the input features are weighted element by element to obtain the weighted features of the h-th branch, as detailed in the following formula.

[0101]

[0102] in: This indicates element-wise multiplication; This is the weighted feature vector output by the h-th branch.

[0103] Furthermore, the outputs of each weighted branch are fused to obtain a comprehensive feature representation, as detailed in the following formula.

[0104]

[0105] in: This is the fusion mapping matrix; The weighted feature vector output by the h-th branch Finally, the fused features The input is then fed into a fully connected neural network for processing to obtain the final system state prediction result, as detailed in the following formula.

[0106]

[0107] Through the aforementioned multi-head feature weighting mechanism, the model can adaptively adjust the weights of each operating parameter under different operating conditions, thereby highlighting the key operating parameters that have a significant impact on the system state and providing a foundation for subsequent identification of key operating parameters.

[0108] To ensure that the prediction results of the power grid state identification model conform to the basic laws of the power system, this embodiment introduces physical constraint loss during model training.

[0109] Physical constraints include node power balance constraints (composed of the active and reactive power of each node), voltage operating range constraints, and line transmission capacity constraints. Correspondingly, physical constraint losses include node power balance losses, node voltage over-limit penalty functions, and line overload penalty functions. These physical constraint losses are characterized by the following formula.

[0110] Model-predicted output Compared with the actual system state Construct a data fitting loss function.

[0111]

[0112] in, This represents the total number of training samples. The first The model prediction output and the actual system state label for each sample.

[0113] To ensure that the model output conforms to the power system operating mechanism, a unified physical constraint loss for power flow is introduced to constrain node power balance, voltage operating range, and line transmission capacity. Based on the power flow equations, power balance residuals are constructed, and the physical constraint loss is uniformly defined as follows, taking into account voltage overruns and line overload conditions.

[0114]

[0115] in, These represent the imbalance of active and reactive power at the node, respectively. The penalty function for exceeding the node voltage limit is defined by the following formula.

[0116]

[0117] The line overload penalty function is defined by the following formula.

[0118]

[0119] in, Node voltage; For line power flow; This refers to the line capacity.

[0120] The data fitting loss and physical constraint loss are weighted and combined to construct the total loss function for model training, as shown in the following formula.

[0121]

[0122] in, These are weighting coefficients used to balance the influence between data fitting and physical constraints.

[0123] By constructing the loss function as described above, while ensuring the model's prediction accuracy, power flow equations and operational constraints are introduced into the model training process. This allows the model to learn data features while satisfying the basic operating laws of the power system, thereby improving the reliability and engineering applicability of the key operating parameter identification results.

[0124] The active and reactive power imbalance at nodes can be calculated based on Kirchhoff's laws, reflecting whether the model's predicted operating state meets the power balance condition. Node power balance loss, by quantifying the sum of squares of the imbalance, forces the model to optimize towards power balance during training. The node voltage over-limit penalty function is constructed similarly to the aforementioned voltage safety index, penalizing voltage predictions that exceed a preset safety range. The line overload penalty function penalizes situations where the line power flow in the model's predicted output exceeds the preset transmission capacity. When the line power flow does not exceed the capacity, the penalty term is 0; once overloaded, the penalty term increases linearly with the degree of overload.

[0125] This embodiment transforms the physical mechanisms of the power system into a mathematical optimization objective using the aforementioned formula. During the backpropagation process of model training, the physical constraint loss acts as a regularization term, working in conjunction with the model prediction loss. If the model predicts operating states that violate physical laws (such as excessive power imbalance or severe voltage exceedances) solely to fit the data, the physical constraint loss will generate a large gradient, thereby correcting the model parameters. This design effectively avoids the illusion phenomenon that may occur with purely data-driven models, ensuring the physical consistency and engineering reliability of the power grid data processing results.

[0126] As one implementation method, based on the above embodiments, the specific process for determining the target operating parameters is described in detail. After the power grid state identification model is trained, the model has the ability to predict system state indicators and output attention weights. However, relying solely on attention weights for key parameter identification may have certain limitations; for example, the model may assign inappropriately high weights to certain redundant features. To further improve the accuracy and robustness of key operating parameter identification, this embodiment introduces a disturbance analysis mechanism, combining the model's internal attention level with the actual degree of external influence to form a more reliable identification result.

[0127] Specifically, this embodiment employs a disturbance analysis-based approach to determine the characteristic contribution of power grid operating parameters across various dimensions. The core idea of ​​disturbance analysis is to quantify the actual impact of a parameter on the system state by artificially altering the value of a specific input parameter and observing the magnitude of the change in the model's output.

[0128] Step S410: Apply a disturbance of a preset amplitude to each dimension of the power grid operating parameter to obtain the power grid operating variable corresponding to each dimension of the power grid operating parameter.

[0129] The preset disturbance magnitude can be set according to actual engineering needs, such as 1%, 5%, or other small increments of the parameter's rated value. It should be understood that the disturbance magnitude should not be too large, lest it cause the system operating point to deviate too far from the current operating condition, leading to the failure of the linearization assumption; nor should it be too small, lest it be overwhelmed by numerical calculation errors. For discrete operating mode parameters, the disturbance can manifest as a state switch, such as switching from line operation to line shutdown. By applying the disturbance, several new sets of input variables are generated, namely, dimensional power grid operating variables.

[0130] Step S420: For any one-dimensional power grid operating parameter, modify the corresponding one-dimensional power grid operating variable among the multiple-dimensional power grid operating variables to obtain the modified multi-dimensional power grid operating variable; input the modified multi-dimensional power grid operating variable into the power grid state identification model to obtain the system state index after disturbance of the one-dimensional power grid operating parameter.

[0131] In practice, the controlled variable method is used, modifying only one dimension's parameter at a time while keeping other dimensions unchanged. For example, to evaluate the impact of the active power load of the i-th load node on the system state, only the active power load value of that node is modified, while the load, generator output, and operating mode of other nodes remain unchanged. The modified multidimensional power grid operation variables are input into the model, which then performs forward calculations and outputs the system state indicators after disturbance. This process simulates the impact of single parameter fluctuations on overall power grid security.

[0132] Step S430: Determine the characteristic contribution of each dimension of the power grid operating parameter based on the difference between the output index of the system state index after the disturbance of each dimension of the power grid operating parameter and the system state index.

[0133] Specifically, the difference in output indicators can be obtained by calculating Euclidean distance, absolute difference, or relative rate of change. If a disturbance to a parameter causes a drastic change in the system state indicators (e.g., a significant increase in voltage safety indicators), it indicates that the parameter is extremely sensitive to system safety and has a high characteristic contribution. Conversely, if the system state indicators show almost no change, it indicates that the parameter has a small impact on the system state under the current operating conditions and has a low characteristic contribution. The characteristic contribution quantifies the degree of objective impact of a change in a single parameter on the system state.

[0134] After obtaining the feature contribution, step S440 determines the comprehensive importance index of each dimension of power grid operation parameters based on the attention weight and feature contribution of each dimension of power grid operation parameters.

[0135] This embodiment constructs a comprehensive importance index by fusing attention weights and feature contribution. Attention weights reflect the strength of the intrinsic correlation between parameters learned by the model based on data-driven learning, representing subjective attention; feature contribution reflects the actual impact of parameter changes on the output, representing objective validation. Combining the two effectively avoids the bias of a single index. For example, if a parameter has a high attention weight but an extremely low feature contribution, it indicates that the model may be overfitting or misjudging, and the actual importance of that parameter should be reduced; conversely, if both are high, then that parameter is confirmed as a key parameter. The comprehensive importance index can be calculated using weighted summation, geometric mean, or by constructing a joint evaluation function; the specific formula can be adjusted according to the actual application scenario.

[0136] Step S450: According to the comprehensive importance index of the power grid operation parameters in each dimension, the power grid operation parameters in each dimension are sorted in descending order. From the power grid operation parameters in the descending order, the power grid operation parameters ranked at the top of the preset percentage threshold are selected as target operation parameters. These target operation parameters are used to guide power grid operation monitoring and dispatching decisions.

[0137] Power grid operating parameters are also called operating parameters.

[0138] An example is the formula for calculating the comprehensive importance index.

[0139]

[0140] in, This represents the comprehensive importance index of the i-th power grid operating parameter; The feature weights corresponding to the power grid operating parameters are calculated by the multi-head feature weighting mechanism described in the above implementation steps, and are used to characterize the degree of influence of each operating parameter on the model prediction results. These are the weighting coefficients.

[0141] Specifically, the descending order ranking places the parameter with the highest overall importance index at the top, decreasing sequentially. Preset percentage thresholds can be set based on the limited availability of scheduling resources and the level of detail required, such as selecting the top 10%, top 20%, or top N parameters. These selected target operating parameters represent the most significant key factors affecting the safe operation of the power grid under the current conditions. Dispatchers can then conduct targeted monitoring and adjustments, such as prioritizing voltage regulation resources for nodes with the highest overall importance index or implementing power flow control on critical lines. This embodiment, through the aforementioned dual verification mechanism, achieves precise identification of key operating parameters, providing reliable data support for refined management and preventative control of the power grid.

[0142] To achieve the above functions, the power grid data processing device includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art will readily recognize that, based on the algorithmic steps of the examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0143] This embodiment provides a power grid data processing device, such as... Figure 2 As shown, this device is used to execute the power grid data processing method of any of the above embodiments. Through a modular design, the device integrates data acquisition, model reasoning, and key parameter identification functions, enabling it to efficiently and accurately extract key information from massive amounts of power grid operation data. Specifically, the device includes an acquisition unit 21 and an identification unit 22.

[0144] The acquisition unit 21 is configured to acquire multi-dimensional power grid operating parameters under various operating conditions; the multi-dimensional power grid operating parameters include the node load parameters, power generation output parameters and operating modes of each node in the power system.

[0145] Specifically, the acquisition unit 21 is the interface module for the device to interact with external data sources. In terms of hardware implementation, the acquisition unit can be a physical component such as a communication interface circuit, a data acquisition card, or a network adapter, establishing a connection with the power system's Supervisory Control and Data Acquisition (SCADA), Wide Area Measurement System (WAMS), or Energy Management System (EMS) via wired or wireless communication. The acquisition unit reads real-time operating data or historical stored data from these systems according to a preset sampling frequency or triggering conditions.

[0146] For example, the acquisition unit 21 can read node voltage and current telemetry data from the substation intelligent electronic device (IED) via the IEC 61850 communication protocol to calculate node load parameters; or it can retrieve generator output plan data and real-time output data from the dispatch center database via a dedicated network interface. Furthermore, the acquisition unit 21 also has data preprocessing capabilities, capable of format conversion, missing value imputation, and outlier removal of the collected raw data, transforming heterogeneous multi-source data into a standardized multi-dimensional power grid operation parameter vector, providing high-quality data input for subsequent model processing.

[0147] It should be understood that the specific implementation of the acquisition unit 21 is not limited to the above example, and any hardware or software module that can realize the data acquisition and transmission function should be covered within the protection scope of this embodiment.

[0148] The identification unit 22 is configured to input multi-dimensional power grid operation data into the power grid state identification model, so that the power grid state identification model first calls the attention mechanism to adaptively weight the power grid operation parameters of each dimension in the multi-dimensional power grid operation parameters to obtain a weighted feature representation; then calls the neural network to determine the system state index of the power system based on the weighted feature representation; wherein, the weighted feature representation is the weighted result based on the power grid operation parameters of each dimension and the attention weights of the power grid operation parameters of each dimension; the attention weight is used to characterize the degree of influence of each power grid operation parameter on the power grid state; the power grid state identification model is trained based on a loss function; the loss function includes physical constraint loss caused by not satisfying physical constraints; and, based on the attention weights of the power grid operation parameters of each dimension, the target operation parameters whose attention weights are higher than the preset weight conditions are determined from the multi-dimensional power grid operation parameters.

[0149] Specifically, the identification unit 22 is the core computing module of the device, which integrates a pre-trained power grid state identification model. In terms of hardware implementation, the identification unit typically consists of a processor (such as a CPU, GPU, TPU, or dedicated AI acceleration chip) and memory. The memory stores computer program code, which, when executed by the processor, implements the various functions of the identification unit.

[0150] The workflow of the identification unit 22 is as follows: First, it receives the multi-dimensional power grid operation parameters output by the acquisition unit 21, loads them into memory, and constructs them into a tensor format that meets the model input requirements. Then, it calls the power grid state identification model for forward inference. Inside the model, the attention mechanism module calculates the attention weights of each dimension parameter through matrix operations. These weights reflect the sensitivity or contribution of the parameter to the system state (such as voltage safety and line load) under the current operating conditions. The identification unit 22 performs a weighted summation of the attention weights and the original parameters to obtain a weighted feature representation highlighting key information. Next, the neural network module (such as fully connected layers, convolutional layers, etc.) performs deep feature extraction and nonlinear mapping on the weighted feature representation, outputting predicted values ​​of the system state indicators. Simultaneously, the identification unit 22 extracts the attention weight vector output by the attention mechanism module and filters the multi-dimensional power grid operation parameters according to a preset screening strategy (such as setting a weight threshold or selecting Top-K sorting). For example, the identification unit 22 can sort all dimensions of operation parameters from largest to smallest according to their attention weights and select the parameters ranked in the top preset percentage threshold (such as the top 10%) as target operation parameters. Finally, the identification unit 22 outputs the determined target operating parameter list to the display device for dispatchers to refer to, or sends it to the automatic control system to execute the corresponding adjustment instructions.

[0151] This embodiment achieves the hardware or software implementation of power grid data processing methods through the collaborative work of the acquisition unit 21 and the identification unit 22, thereby improving the efficiency and automation level of data processing and providing strong technical support for the intelligent operation and monitoring of the power grid.

[0152] Regarding the apparatus in the above embodiments, the specific manner in which each unit module performs its operations has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0153] This embodiment provides a power system. This power system aims to implement the power grid data processing method described in the above embodiments through hardware architecture integration, thereby enabling the identification and decision support of key operating parameters in a real physical power grid environment. The power system includes a processor, multiple generators, new energy equipment, and load equipment. Each load equipment corresponds to a node, which stores a computer program. When the program is executed by the processor, it implements the power grid data processing method according to any one of the above embodiments.

[0154] Figure 3 This is a schematic diagram of a control device provided in this application. Figure 3The control device 50 may include at least one first processor 501 and a memory 503 for storing processor-executable instructions. The first processor 501 is configured to execute the instructions in the memory 503 to implement the power grid data processing method in the following embodiments.

[0155] In addition, the control device 50 may also include a communication bus 502, at least one communication interface 504, an input device 506, and an output device 505.

[0156] The first processor 501 may be a processor (central processing unit, CPU), a microprocessor unit, an ASIC, or one or more integrated circuits for controlling the execution of programs according to the present application.

[0157] The communication bus 502 may include a path for transmitting information between the aforementioned components.

[0158] Communication interface 504 uses any transceiver-like device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0159] Input device 506 is used to receive input signals and output device 505 is used to output signals.

[0160] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed discs, laser discs, optical discs, digital universal discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory may exist independently and be connected to the processing unit via a bus. Memory may also be integrated with the processing unit.

[0161] The memory 503 stores instructions for executing the scheme of this application, and the execution is controlled by the first processor 501. The first processor 501 executes the instructions stored in the memory 503 to realize the functions of the method of this application.

[0162] In a specific implementation, as one example, the first processor 501 may include one or more CPUs, for example... Figure 3 CPU0 and CPU1 in the CPU.

[0163] In a specific implementation, as one example, the control device 50 may include multiple processors, such as... Figure 3 The first processor 501 and the second processor 507 are described. Each of these processors can be a single-core processor or a multi-core processor. A processor here can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0164] The control device, such as Figure 3 The diagram shows a first processor 501 and a memory 503 for storing executable instructions of the first processor 501. The first processor 501 is configured to execute the executable instructions to implement the power grid data processing method as described in any of the possible embodiments above. Since the same technical effects can be achieved, further details are omitted here to avoid repetition.

[0165] This application also provides a computer-readable storage medium, which, when the instructions in the computer-readable storage medium are executed by the processor of a power grid data processing device or control device, enables the power grid data processing device or control device to perform a power grid data processing method as described in any of the above possible embodiments. And it can achieve the same technical effect; to avoid repetition, it will not be described again here.

[0166] This application also provides a computer program product, including a computer program or instructions, which are executed by a processor as a power grid data processing method according to any of the possible implementations described above. Furthermore, it achieves the same technical effects, and to avoid repetition, it will not be described again here.

[0167] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0168] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A power grid data processing method, characterized in that, The method includes: The system acquires multi-dimensional power grid operating parameters under various operating conditions; these parameters include node load parameters, power generation output parameters, and operating modes of each node in the power system. The multidimensional power grid operation data is input into the power grid state identification model. The model first invokes an attention mechanism to adaptively weight the power grid operation parameters in each dimension of the multidimensional power grid operation parameters, obtaining a weighted feature representation. Then, a neural network is invoked to determine the system state indicators of the power system based on the weighted feature representation. The weighted feature representation is a weighted result based on the power grid operation parameters in each dimension and their respective attention weights. The attention weights characterize the degree of influence of each power grid operation parameter on the power grid state. The power grid state identification model is trained based on a loss function, which includes physical constraint losses caused by non-compliance with physical constraints. Furthermore, based on the attention weights of the power grid operating parameters in each dimension, target operating parameters with attention weights higher than preset weight conditions are determined from the multi-dimensional power grid operating parameters.

2. The method according to claim 1, characterized in that, The method further includes: The system acquires multidimensional historical operating parameters of the power system under various operating conditions, using these parameters as input samples. It also acquires historical system state indicators of the power system under each operating condition, using these indicators as output samples corresponding to the input samples. Based on the input sample, construct an input feature vector, and based on the output sample corresponding to the input sample, construct an output feature vector corresponding to the input feature vector; Based on the input feature vector, the model parameters of the initial identification model are adjusted and trained until the loss function reaches the preset loss condition. The initial identification model corresponding to the model parameters when the loss function reaches the preset loss condition is determined as the power grid state identification model. The loss function also includes the model prediction loss between the output prediction index representing the system state index of the initial identification model and the corresponding output feature vector.

3. The method according to claim 2, characterized in that, Both the historical system status indicators and the system status indicators include at least the following indicators: voltage safety indicators and line load indicators; The input feature vector x is represented by the following formula; Among them, the power system includes load nodes and One generator, , The first Under the first operating condition Active and reactive loads of each load node; , The first Under the first operating condition The active and reactive power outputs of the generator nodes. For the first The operating mode feature vector corresponding to each operating condition; , , The first Load level coefficient, renewable energy output ratio, and tie line power level coefficient under each operating condition; Input feature vector under various operating conditions ; Voltage safety indicators are characterized by the following formula; in, Indicates the first Voltage safety indicators under various operating conditions; Indicates the first Under the first operating condition Voltage amplitude at each node; , They represent the first The upper and lower limits of the allowable voltage for each node; The line load index is represented by the following formula: in, Indicates the first Line load indicators under various operating conditions; Indicates the total number of lines in the system; Indicates the first The allowable transmission capacity of each line; Indicates the first Under the first operating condition The power flow value of the line.

4. The method according to claim 2, characterized in that, The acquisition of historical system state indicators of the power system under multi-dimensional historical operating parameters for each operating condition includes: Based on the node admittance matrix, node injected power and network topology of the power system under each operating condition and corresponding multidimensional historical operating parameters, power flow calculation is performed to obtain the historical voltage amplitude and historical line power flow of each node under each operating condition and corresponding multidimensional historical operating parameters. Based on the historical voltage amplitude of each node under each operating condition and the corresponding multi-dimensional historical operating parameters, determine the historical voltage safety index of each node under each operating condition and the corresponding multi-dimensional historical operating parameters. The proportion of historical line load to allowable transmission capacity for each node under each operating condition and corresponding multi-dimensional historical operating parameters is determined as the historical line load index for each node under each operating condition and corresponding multi-dimensional historical operating parameters.

5. The method according to any one of claims 1 to 4, characterized in that, The physical constraints include node power balance constraints (composed of the active and reactive power of each node), voltage operating range constraints, and line transmission capacity constraints. The physical constraint losses include node power balance losses, node voltage over-limit penalty functions, and line overload penalty functions. The node power balance losses are determined based on the imbalance between the active and reactive power of each node. The node voltage over-limit penalty function is used to penalize situations where the voltage in the model prediction output exceeds a preset safe range. The line overload penalty function is used to penalize situations where the line power flow in the model prediction output exceeds a preset transmission capacity. The physical constraint loss is characterized by the following formula; in, These represent the imbalance of active and reactive power at the node, respectively. This represents the penalty function for exceeding the node voltage limit. This represents the line overload penalty function. Indicates the first Under the first operating condition Voltage amplitude at each node; , They represent the first The upper and lower limits of the allowable voltage for each node; Node voltage; For line power flow; This refers to the line capacity.

6. The method according to any one of claims 1 to 4, characterized in that, The step of determining target operating parameters with attention weights higher than preset weight conditions from the multi-dimensional power grid operating parameters based on the attention weights of the power grid operating parameters in each dimension includes: Based on disturbance analysis, the characteristic contribution of power grid operating parameters in various dimensions is determined. Based on the attention weight and characteristic contribution of each dimension of power grid operation parameters, a comprehensive importance index for each dimension of power grid operation parameters is determined. Based on the comprehensive importance index of the power grid operation parameters in each dimension, the power grid operation parameters in each dimension are arranged in descending order; From multiple power grid operation parameters arranged in descending order, the power grid operation parameters that rank at the top of a preset percentage threshold are selected as the target operation parameters. The target operation parameters are used to guide power grid operation monitoring and dispatch decisions.

7. The method according to claim 6, characterized in that, The method based on disturbance analysis determines the characteristic contribution of power grid operating parameters in various dimensions, including: A perturbation of a preset magnitude is applied to each dimension of the power grid operating parameter to obtain the power grid operating variable corresponding to each dimension of the power grid operating parameter; For any one-dimensional power grid operating parameter, modify the corresponding one-dimensional power grid operating variable among the multiple-dimensional power grid operating variables to obtain the modified multi-dimensional power grid operating variable; input the modified multi-dimensional power grid operating variable into the power grid state identification model to obtain the system state index after the disturbance of the one-dimensional power grid operating parameter. The characteristic contribution of each dimension of the power grid operating parameter is determined based on the difference between the output index of the system state index after the disturbance of each dimension of the power grid operating parameter and the system state index.

8. A power grid data processing device, characterized in that, The device includes: The acquisition unit is configured to acquire multi-dimensional power grid operating parameters under various operating conditions; the multi-dimensional power grid operating parameters include the node load parameters, power generation output parameters, and operating modes of each node in the power system. The identification unit is configured to input the multidimensional power grid operation data into a power grid state identification model, so that the power grid state identification model first invokes an attention mechanism to adaptively weight the power grid operation parameters of each dimension in the multidimensional power grid operation parameters to obtain a weighted feature representation; then invokes a neural network to determine the system state index of the power system based on the weighted feature representation; wherein, the weighted feature representation is a weighted result based on the power grid operation parameters of each dimension and the attention weights of the power grid operation parameters of each dimension; the attention weights are used to characterize the degree of influence of each power grid operation parameter on the power grid state; the power grid state identification model is trained based on a loss function; the loss function includes physical constraint loss caused by not satisfying physical constraints; and, based on the attention weights of the power grid operation parameters of each dimension, determines the target operation parameters from the multidimensional power grid operation parameters whose attention weights are higher than a preset weight condition.

9. An electric power system, characterized in that, It includes a processor, multiple generators, new energy equipment and load equipment, with one load equipment corresponding to one node, on which a computer program is stored. When the program is executed by the processor, it implements the power grid data processing method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the power grid data processing method according to any one of claims 1 to 7.

Images