New energy declaration data multi-source checking method based on GAT-CKDE algorithm

Abstract

The application discloses a new energy declaration data multi-source checking method based on a GAT-CKDE algorithm, and comprises the following steps: identifying abnormal data and missing data of original new energy declaration data through a pre-constructed composite abnormality judgment function, and outputting marked abnormal data, wherein the abnormal data comprises abnormal data and missing data; repairing the abnormal data by using a pre-constructed graph attention network structure to obtain repaired new energy prediction data; and calculating a risk envelope of the repaired new energy prediction data by using a pre-constructed new energy prediction data risk envelope generation method based on a conditional kernel density estimation.

Description

Technical Field

[0001] This invention relates to the field of power system dispatching technology, and more specifically, to a multi-source verification method for new energy application data based on the GAT-CKDE algorithm. Background Technology

[0002] The randomness and volatility of renewable energy generation necessitate reliance on forecasting information for bidding and dispatching decisions in the electricity market. Especially in a multi-timescale spot market system, renewable energy generation companies must forecast and submit transaction applications at different timescales, including day-ahead, intraday, and real-time markets. The accuracy and responsiveness of these forecasts directly impact market revenue, system operational safety, and the level of renewable energy absorption. When renewable energy plants submit planned power generation and output capacity to the dispatching agency, they may employ strategic considerations based on profit maximization, such as overestimating their short-term power generation potential to secure more grid-connected power quotas and thus increase operating profits. As the core department ensuring the safe and stable operation of the power grid, the dispatching agency must reasonably verify the output declaration data from renewable energy plants. This verification is not only crucial for maintaining a fair order in the electricity market but also a necessary means to prevent safety risks such as grid frequency fluctuations, voltage instability, or localized supply-demand imbalances caused by excessive discrepancies between actual output and declared output.

[0003] Currently, typical overseas markets have formed a verification system based on technological empowerment, economic incentives, and cross-regional collaboration. The US PJM market has built a spatiotemporal hierarchical management mechanism covering ultra-short-term to long-term forecasts. It improves anomaly identification accuracy through machine learning models such as LSTM, and combines multi-timescale verification with a deviation margin system, strongly linking forecast accuracy to market returns. A forecast accuracy below 80% triggers an assessment. The California CAISO market mandates new energy sources to connect to the VER forecasting system, employing a real-time scheduling mechanism updated every 5 minutes. It guides entities to improve forecasting capabilities by settling deviations in electricity consumption at real-time electricity prices, and allows flexible adjustments to declared quantities and forecast values ​​to reduce risk. European markets emphasize cross-regional collaboration. Nord Pool in Northern Europe integrates cross-border new energy forecasting data through a power flow coupling mechanism, using the VPP platform to aggregate distributed resources and optimize forecast accuracy. The German EPEXSPOT market introduces fundamental models to dynamically adjust forecast curve weights to adapt to extreme weather scenarios. The Australian AEMO adopts a dual-source verification model of self-reported and official forecasts, dynamically determining scheduling criteria based on historical accuracy, and promotes a self-reporting certification mechanism, allowing certified entities to participate in rapid frequency response services. Furthermore, internationally, quantitative anomaly detection thresholds and tiered replacement rules have been widely established, combined with blockchain technology to achieve data storage and traceability, ensuring the credibility of verification. Markets in Europe and the US, represented by CAISO in California, PJM in Pennsylvania, New Jersey, and Maryland, and EPEX in Europe, have set clear and quantitative anomaly detection thresholds. These thresholds are typically linked to the nameplate capacity of generator sets or recent historical performance, such as CAISO's "exceeding three times the nameplate capacity" or PJM's "exceeding twice the maximum value in the past 30 days," providing objective standards for automatic anomaly identification. Regarding data replacement, a tiered replacement strategy is generally adopted, following the principle of "from near to far, from physical to virtual": the first priority is redundant meters, which directly obtain data using physical backup equipment, offering the highest reliability. The second priority is historical data fitting. When there are no redundant meters, the average value of historical data from the same period (e.g., the past 7 days, 30 days, or 14 days) under "same time period" or "similar temperature" conditions is used as a replacement to approximate the actual situation as closely as possible. In terms of settlement and retrospective analysis, these markets demonstrate strong flexibility. Settlement may use time-of-use pricing or average prices. Their retrospective recalculation mechanism allows for the recalculation of electricity charges using corrected data after the fact (ranging from T+55 days to a maximum of 3 years), ensuring the accuracy of medium- and long-term settlements and effectively correcting injustices caused by data issues. The Nord Pool in Northern Europe and the Elexon market in the UK place greater emphasis on procedural fairness and verification between market participants in their processing logic. They typically do not pre-set fixed numerical thresholds; instead, when data is in doubt, the transmission system operator (TSO) or settlement institution first communicates and verifies with the power generation entity. Only after jointly confirming that the data is "invalid" is the replacement procedure initiated. Their replacement method is also highly distinctive, heavily relying on the planned or declared data submitted by the market participants themselves.For example, Nord Pool prioritizes using "planned reported values" for adjustments, while the UK BM market directly uses its "balancing market reported volume" to replace short-term gaps. This approach closely links data responsibility with market responsibility, incentivizing participants to provide accurate and reliable forecasts and plans. Regarding backtesting mechanisms, the UK utilizes a corrective settlement mechanism, while the Nordic countries use the TSO for unified processing during monthly balancing. Both have relatively long backtesting periods (the UK's final settlement correction is typically completed 16-28 months after the transaction) and introduce a financial interest compensation mechanism, which, while making up for the shortfall, also considers the time value of money, reflecting refined settlement management.

[0004] Existing tools for verifying forecast data are inefficient. For example, rigid measures such as fixed thresholds restrict the self-declaration of renewable energy entities, resulting in a large number of missing and outlier data in the declaration data received by the dispatch center, which are difficult to identify and correct. Furthermore, the processing relies on manual labor or simple algorithms, which cannot guarantee the quality of the repaired data, leading to increased dispatch costs or reduced power system security. Therefore, a feedforward, proactive intelligent tool is needed to transform renewable energy declaration data into accurate, reliable, and high-quality dispatch decision-making data that includes quantifiable risk information, thereby achieving global economic optimization of system operation while ensuring grid security. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a multi-source verification method for new energy application data based on the GAT-CKDE algorithm.

[0006] According to one aspect of the present invention, a multi-source verification method for new energy application data based on the GAT-CKDE algorithm is provided, comprising: The original new energy application data is identified as abnormal and missing by a pre-constructed composite anomaly detection function, and the abnormal data is marked as abnormal data and missing data. A pre-built graph attention network structure is used to repair abnormal data, resulting in repaired new energy prediction data. A pre-constructed risk envelope generation method based on conditional kernel density estimation for new energy prediction data is used to calculate and repair the risk envelope of new energy prediction data.

[0007] Optionally, the construction steps of the composite anomaly determination function include: Construct regular anomaly detection functions, logical consistency detection functions, and missing item detection functions; A composite anomaly detection function is constructed based on the regular anomaly detection function, the logical consistency detection function, and the missing information detection function.

[0008] Optionally, the construction steps of the regular anomaly detection function include: Construct functions for determining limits, mutations, and negative values; Based on the over-limit judgment function, the mutation judgment function, and the negative value judgment function, a regular anomaly judgment function is constructed. The expression for the over-limit determination function is:

[0009] In the formula, For new energy power stations In time The declared data, This is the sum of the nameplate values ​​of the equipment working at the station. Configurable parameters are set according to different regulatory rules; The expression for the mutation determination function is:

[0010] In the formula, From Time's up Output dataset under similar weather conditions at any given time; These are configurable parameters set according to different regulatory rules; The expression for the negative value determination function is:

[0011] The expression for the regular anomaly detection function is:

[0012] when When the output value is 1, the data point is marked as a regularity anomaly; when... When the output value is 0, the data point is considered to be normal data with regularity.

[0013] Optionally, the expression for the logical consistency determination function is:

[0014] In the formula, For new energy power stations Location and time solar altitude angle, An extremely small positive threshold is set to mitigate sensor noise or minute measurement errors. This is an indicator sign for new energy types. Represents new energy For photovoltaic power stations, Represents new energy For wind farms, For process-level monitoring of SCADA data, For settlement-level metering TMR data, The maximum allowable relative deviation threshold; when When the value is 1, the data point is marked as having a logical consistency anomaly; when... When the value is 0, the data point is considered to have normal logical consistency. The expression for the missing value determination function is: If data is missing, record ,otherwise ; The expression for the composite anomaly detection function is: .

[0015] Optionally, a pre-built graph attention network structure is used to repair the abnormal data, resulting in repaired new energy prediction data, including: All new energy power stations within the dispatch area are abstracted into a graph structure, resulting in a graph network structure; The initial edge weights are calculated based on the graph network structure, and the original adjacency matrix is ​​constructed based on the initial edge weights. The original adjacency matrix is ​​subjected to self-looping and symmetric normalization to obtain a normalized adjacency matrix; Scan the feature vectors of all nodes. If the data quality mask bit in the feature vector is set to an abnormal state, then mark the node as the target node for repair. Taking the target node to be repaired as the center, extract its local subgraph, which contains the target node to be repaired and all its neighboring nodes; The local subgraph is input into the graph attention network model, and the attention weight of each neighbor node to the target node to be repaired is dynamically calculated. Based on attention weights, the feature information of neighboring nodes is weighted and aggregated to infer the missing or abnormal data of the target node to obtain the new energy prediction data for repair.

[0016] Optionally, a pre-built risk envelope generation method for new energy prediction data based on conditional kernel density estimation is used to calculate and repair the risk envelope of the new energy prediction data, including: Based on historical actual output values ​​and historical predicted values, the prediction error at each historical time point is calculated to form a dataset containing historical predicted values ​​and historical prediction errors. To correct the current forecast values ​​in the new energy forecast data, by setting the window width, all historical forecast errors that meet the preset conditions are filtered out from the dataset to form a set of similar historical errors; Conditional kernel density estimation is applied to error data in similar historical error sets to estimate the error probability density function under the current prediction conditions; The cumulative distribution function is calculated based on the conditional error probability density function, and the error quantile corresponding to the target confidence level is determined. Combined with the current forecast value, a risk envelope for repairing new energy forecast data is generated.

[0017] According to another aspect of the present invention, a multi-source verification device for new energy application data based on the GAT-CKDE algorithm is provided, comprising: The identification module is used to identify abnormal and missing data in the original new energy application data through a pre-built composite anomaly judgment function, and output marked abnormal data, which includes abnormal data and missing data. The repair module is used to repair abnormal data using a pre-built graph attention network structure to obtain repaired new energy prediction data. The calculation module is used to calculate the risk envelope of the repaired new energy prediction data using a pre-built risk envelope generation method based on conditional kernel density estimation.

[0018] According to another aspect of the present invention, a computer-readable storage medium is provided, the storage medium storing a computer program for performing the methods described in any of the above aspects of the present invention.

[0019] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the instructions to implement the method described in any of the preceding aspects of the present invention.

[0020] Therefore, the multi-source verification method for new energy application data based on the GAT-CKDE algorithm proposed in this invention provides high-precision new energy output prediction data for the power grid dispatch center by constructing a verification system that integrates intelligent algorithms. This enables dispatch decision-making to shift from passively responding to deviations to actively managing uncertainty, significantly reducing the system's reserve and balancing costs and improving the effectiveness of dispatch tools. Simultaneously, the method proposed in this patent overcomes the problems of low efficiency and high human resource consumption of manual correction through an automated verification mechanism, helping to reduce the labor costs of verifying new energy application data. Ultimately, while ensuring power grid security, it achieves the optimization of total system operating costs and a dynamic balance between system operating efficiency and security. Attached Figure Description

[0021] Exemplary embodiments of the present invention can be more fully understood by referring to the following figures: Figure 1 This is a flowchart illustrating a multi-source verification method for new energy application data based on the GAT-CKDE algorithm, provided in an exemplary embodiment of the present invention. Figure 2 This is a flowchart illustrating a multi-source verification method for new energy application data based on the GAT-CKDE algorithm, provided in an exemplary embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of a multi-source verification device for new energy application data based on the GAT-CKDE algorithm provided in an exemplary embodiment of the present invention; Figure 4 This is the structure of an electronic device provided in an exemplary embodiment of the present invention. Detailed Implementation

[0022] Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0023] It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the invention.

[0024] Those skilled in the art will understand that the terms "first," "second," etc., in the embodiments of the present invention are only used to distinguish different steps, devices, or modules, and do not represent any specific technical meaning, nor do they indicate a necessary logical order between them.

[0025] It should also be understood that in the embodiments of the present invention, "multiple" can refer to two or more, and "at least one" can refer to one, two or more.

[0026] It should also be understood that any component, data or structure mentioned in the embodiments of the present invention can generally be understood as one or more unless explicitly defined or given contrary instructions in the context.

[0027] Furthermore, the term "and / or" in this invention is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this invention generally indicates that the preceding and following related objects have an "or" relationship.

[0028] It should also be understood that the description of the various embodiments in this invention emphasizes the differences between the various embodiments, and the similarities or similarities can be referred to each other. For the sake of brevity, they will not be described in detail.

[0029] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0030] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0031] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, they should be considered part of the specification.

[0032] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0033] The embodiments of this invention can be applied to electronic devices such as terminal devices, computer systems, and servers, and can operate together with a wide range of other general-purpose or special-purpose computing system environments or configurations. Well-known examples of terminal devices, computing systems, environments, and / or configurations suitable for use with electronic devices such as terminal devices, computer systems, and servers include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputer systems, mainframe computer systems, and distributed cloud computing environments including any of the above systems, etc.

[0034] Electronic devices such as terminal devices, computer systems, and servers can be described in the general context of computer system executable instructions (such as program modules) executed by a computer system. Typically, program modules can include routines, programs, object programs, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. Computer systems / servers can be implemented in distributed cloud computing environments, where tasks are executed by remote processing devices linked through communication networks. In distributed cloud computing environments, program modules can reside on local or remote computing system storage media, including storage devices.

[0035] Exemplary method Figure 1 This is a flowchart illustrating a multi-source verification method for new energy application data based on the GAT-CKDE algorithm, provided in an exemplary embodiment of the present invention. This embodiment can be applied to electronic devices, such as… Figure 1 As shown, the multi-source verification method 100 for new energy declaration data based on the GAT-CKDE algorithm includes the following steps: Step 101: Identify abnormal and missing data in the original new energy application data using a pre-constructed composite anomaly judgment function, and output marked abnormal data, where abnormal data includes abnormal data and missing data; Step 102: The abnormal data is repaired using a pre-constructed graph attention network structure to obtain repaired new energy prediction data; Step 103: Calculate and repair the risk envelope of the new energy prediction data using a pre-built risk envelope generation method based on conditional kernel density estimation for new energy prediction data.

[0036] This invention proposes a multi-source verification mechanism for new energy application data based on the GAT-CKDE algorithm, and its implementation process includes the following detailed steps: Step 1: Proactive identification and labeling of anomalies and missing data in new energy forecasting: (1) Regular anomaly identification and marking Hard threshold rules include upper limits on nameplate values, monthly maximum values, and negative value limits. When new energy forecast data exceeds a certain multiple of the nameplate value, it is judged as abnormal - out-of-limit data, and an out-of-limit judgment function is used. The notation is as shown in equation (1): (1) In the formula, For new energy power stations In time The declared data, This is the sum of the nameplate values ​​of the equipment working at the station. These are configurable parameters set according to different regulatory rules. For example, the US CAISO stipulates that the reported forecast data must not exceed 3 times the nameplate capacity, while the UK market stipulates that it must not exceed 1.2 times the nameplate capacity.

[0037] When the new energy forecast data exceeds a certain multiple of the maximum value over a past period, it is judged as abnormal - abrupt data, and the abrupt change judgment function is used. The notation is as shown in equation (2): (2) In the formula, These are configurable parameters set according to different regulatory rules. For example, the US PJM market requires that the reported forecast data should not exceed twice the maximum value in the past 30 days.

[0038] These are configurable parameters set according to different regulatory rules. The criteria for determining similar weather days are shown in Table 1.

[0039] Table 1 Reference Standards for Judging Similar Weather Days

[0040] All meteorological data are derived from numerical weather prediction systems or on-site meteorological monitoring equipment, and data temporal synchronization and spatial consistency must be guaranteed. If a historical period meets the above necessary conditions and also meets ≥4 of the non-necessary conditions, it is determined to be a "similar weather day," and its output data can be used for maximum value comparison in regular anomaly identification.

[0041] When the predicted value is less than zero, it is judged as outlier - negative data, and the negative value judgment function is used. The notation is as shown in equation (3): (3) Regular anomaly detection function Screening for outliers according to rules is performed, as shown in equation (4): (4) when When the output value is 1, the data point is marked as a "regularity anomaly". When the output value is 0, the data point is considered "normal". This function is scalable; for example, by adding criteria for other abnormal regularities such as long-term interruptions, frequent corrections, and exceeding the ramp rate limit, the function can be modified as shown in equation (5). (5) In the formula, This represents the number of types of regular anomalies.

[0042] (2) Logical consistency anomaly identification and marking Logical consistency rules include the photovoltaic nighttime zeroing rule and the dual-source data consistency rule, which can be determined through a logical consistency determination function. Perform logical outlier screening, as shown in equation (6): (6) In the formula, For new energy power stations Location and time solar altitude angle, An extremely small positive threshold is set to mitigate sensor noise or minute measurement errors. This is an indicator sign for new energy types. Represents new energy For photovoltaic power stations, Represents new energy For wind farms, For process-level monitoring of SCADA data, For settlement-level metering TMR data, This is the maximum permissible relative deviation threshold.

[0043] when When the value is 1, the data point is marked as "logical consistency anomaly". When the value is 0, the data point is considered to have "normal logical consistency".

[0044] (3) Missing data processing and labeling When no data is read, it is padded with zero values ​​and marked as missing. Let the missing data determination function be... If data is missing, record .

[0045] (4) Composite anomaly detection function The composite anomaly detection function is determined by combining the above three rules. As shown in equation (7): (7) When data is identified as anomalous, the corresponding data quality mask bit is immediately set to 1.

[0046] Step 2: Spatial correlation anomaly data repair based on graph attention mechanism (GAT); Spatial correlation anomaly data repair aims to correct rule-labeled anomalous data. This invention employs a graph attention network to dynamically capture the complex spatial correlations between stations that vary with meteorological conditions, achieving high-precision intelligent data repair based on dynamic spatial correlations.

[0047] (1) Construction of the spatial topology map of power grid new energy power stations All new energy power stations (such as wind farms and photovoltaic power stations) within the dispatch area are abstracted into a graph structure. In the formula As a set of nodes, new energy power stations Corresponding node . Let be a set of edges, nodes and The connecting edge corresponds to the edge This characterizes its spatial correlation. The construction of edges is based on multiple factors, including geographical proximity, meteorological homogeneity, and electrical correlation. For a set of feature vectors, the elements To represent nodes At any moment eigenvalues A 3D feature vector. Feature values ​​include the station's operational data, meteorological data, installed capacity, and technology type (wind and solar power generation types are encoded as data), etc.

[0048] (2) Construction of spatially related edges and normalized adjacency matrix To enable GTA to effectively utilize geographical and power grid structure information, this invention considers power grid connectivity, geographical distance, and statistical correlation between nodes to construct composite edge weights. If nodes... Corresponding new energy power stations If they are directly connected on the same busbar, in the same substation, or through a transmission line, then the nodes Inter-grid connectivity edge weight Set to 1 otherwise set to 0, as shown in equation (8): (8) Calculate geographical distance ,node Geographical distance between the boundary weights As shown in equation (9): (9) In the formula, This is the distance attenuation parameter.

[0049] compute nodes Historical output Pearson correlation coefficient ,node Statistical correlation between the side weights As shown in equation (10): (10) In the formula, The threshold for statistical correlation is usually set between 0.6 and 0.8.

[0050] node Time series Pearson correlation coefficient The calculation method is shown in equation (11): (11) In the formula, For nodes Time series Pearson correlation coefficient For history Average output of stations within a day It can be determined based on the actual situation, but it is generally taken as 30 days.

[0051] Based on the above information, the initial edge weight calculation method can be defined as shown in equations (12) and (13): (12) (13) In the formula, For nodes The right to the border, The weighting coefficient can be determined based on expert recommendations or historical data.

[0052] Let the number of new energy nodes be Then the original adjacency matrix As shown in equation (14) matrix: (14) To stabilize the training of a graph neural network, the neighbor edge weights need to be symmetrically normalized. This requires a self-looping adjacency matrix. As shown in equation (15): (15) Degree matrix As shown in equation (16): (16) Degree matrix of inverse square root As shown in equation (17): (17) Normalized adjacency matrix and its elements As shown in equations (18) and (19): (18) (19) (3) GAT repair process starts automatically The system scans the feature vectors of all nodes. If the feature vectors... If the data quality mask bit in the node is set to 1, then the node is marked as a repair target, triggering the GAT repair process. (Based on node...) Centered on the target node, a local subgraph including its neighboring nodes is extracted and input into the GAT model for repair. The neighboring node is defined as a node whose edge weight with the target node is not 0. For details on the determination method, please refer to sub-step (2).

[0053] (4) Neighbor node information aggregation and reasoning based on graph attention mechanism The purpose of this step is to dynamically calculate the influence weights of neighboring nodes on the target node using GAT, and to aggregate information in a weighted manner to infer missing values. The attention coefficient physically represents the association strength under the current feature, characterizing the neighboring nodes. For the target node The importance of is shown in equation (20) (for ease of representation, the following discussion is all at the same time). Subscripts omitted in formula ): (20) In the formula, For neighboring nodes For the target node At any moment Attention coefficient. The target node and its neighboring nodes are at the 1st The feature vector of the layer. It is a learnable weight matrix used to perform linear transformations on node features, thereby improving expressive power. It is the transformed feature dimension. This is a learnable attention weight vector used to parameterize the attention mechanism. It is a non-linear activation function, defined as in equation (21), which allows small gradients for negative inputs and avoids the death of neurons in the training model. The pointer concatenation operation results in the following column vector having a different number of rows: .

[0054] (twenty one) Attention weights are obtained by normalizing the attention coefficients using the Softmax function. As shown in equation (22): (twenty two) In the formula, It is an exponential function, ensuring that all coefficients are positive. For the target node The set of neighboring nodes. Representing neighboring nodes For the target node The relative influence weights. For example, when the prevailing wind direction is northwest, the weight of the northwest neighbor will automatically increase.

[0055] Based on the attention weights, the transformation features of the neighboring nodes are weighted and summed, and the updated features of the target node are obtained through the ELU function, as shown in Equation (23): (twenty three) In the formula, After updating the target node (the first one) The feature vector of the layer), ELU is an exponential linear unit, defined as shown in equation (24). The ELU function can alleviate the gradient vanishing problem when multiple layers of GAT are stacked, and ensure that the parameters of deep GAT can be effectively updated.

[0056] (twenty four) To make the learning process more stable and the expressive ability stronger, this invention preferably adopts a multi-head attention mechanism, in which each "head" uses independent parameters. , , Different attention heads are used to guide different learning focus patterns through different loss functions. For example, head 1 focuses on reducing the repair error of geographically neighboring nodes, head 2 focuses on reducing the repair error of meteorologically similar nodes, and head 3 focuses on reducing the repair error of electrically connected nodes. Let the number of attention heads be... K ,available K The updated feature vectors are averaged, as shown in Equation (25): (25) In practical applications, small-scale networks can use 1-2 layers of GAT to accurately capture the spatial dependencies between sites; for scenarios with more than 200 nodes or complex relationships, 3 layers can be considered to alleviate gradient vanishing.

[0057] Step 2 ultimately yields the set of feature vectors after outlier repair.

[0058] Step 3: A method for generating risk envelopes for new energy prediction data based on conditional kernel density estimation; After identification and correction in steps 1 and 2, a relatively accurate prediction data curve can be obtained, but uncertainty still exists. The risk envelope expands a single prediction point into a probability interval, intuitively quantifying the uncertainty risk of the prediction result. This quantified risk can help the power grid dispatch backup power more economically and safely, and reduce operational risks.

[0059] (1) Collect historical prediction error data and screen similar historical errors Historical time points The prediction error is calculated as shown in equation (26): (26) In the formula, The value represents the prediction error; a positive value indicates an underestimation, and a negative value indicates an overestimation. For time The actual output value. For time The predicted value at that time.

[0060] It can be obtained by including This refers to a dataset of historical predicted values ​​and historical prediction errors. Based on the current predicted value, historical errors under similar conditions can be filtered out, defining a model based on the current prediction. The set of similar historical errors selected is As shown in equation (27): (27) In the formula, This is the window width; this value is adjustable.

[0061] For sets In Error data Conditional kernel density estimation is used to estimate the probability density function of the error under that condition. As shown in equation (28): (28) In the formula, For set The number of samples in the sample. Bandwidth controls the smoothness of the density curve. A smaller value results in a smoother curve. It will produce a curve with large fluctuations, a larger This will produce a smoother curve. Bandwidth can usually be determined through cross-validation or rules of thumb (such as Silverman's rule of thumb). For the kernel function, this invention selects the Gaussian kernel, which performs well with high-dimensional and complex distributed data, as shown in equation (29): (29) (3) Calculate the cumulative distribution function (CDF) and generate the risk envelope. From probability density function Calculate the cumulative distribution function As shown in equation (30): (30) To construct the 90% prediction interval (i.e., the risk envelope), we need to find the 5% and 95% quantiles: lower quantile Seeking a solution

[0062] Upper quantile Seeking a solution

[0063] Risk envelope upper and lower bounds , As shown in equation (31): (31) This step enables refined modeling of the uncertainty in renewable energy output forecasting. Compared to traditional verification methods using fixed intervals, this mechanism dynamically constructs a conditional error distribution based on the current forecast value, adaptively generating a risk envelope with a clear statistical confidence level. This overcomes the conservative or overly aggressive problems caused by the fixed interval method neglecting the dynamic characteristics of the forecast value. Furthermore, by introducing conditional distribution modeling, it significantly improves the accuracy and scenario adaptability of risk quantification. This risk envelope provides a probabilistic decision-making basis for grid dispatching and operation, enabling the scientific allocation of reserve capacity and optimization of real-time balancing strategies under the premise of strictly meeting predefined confidence levels. This fundamentally enhances the robustness of the power system to renewable energy uncertainties, achieving synergistic optimization of safety and economy.

[0064] To further understand this invention, the following example of a regional new energy cluster in a province in eastern my country will be used to explain the practical application of this invention.

[0065] This case study is based on operational data from a provincial power grid in eastern China in May 2024. An observation day with typical complex meteorological conditions was selected, and the reported data from 08:00 to 16:00 on that day was analyzed. The weather on that day was cloudy turning sunny with sudden changes in wind direction, leading to significant fluctuations in renewable energy output. The case study covers the day-ahead reported data from 30 renewable energy stations received by the dispatch center, including 15 photovoltaic power stations and 10 wind farms, with a total installed capacity of 2150MW, of which 1183.57MW is photovoltaic and 966.43MW is wind power. To accurately evaluate the anomaly identification effect, the actual output data for that day was obtained as a benchmark. The data shows that due to rapid weather changes, there are two typical problems in the reported data: first, prediction deviations caused by sudden changes in cloud conditions at photovoltaic power stations; and second, prediction inaccuracies at wind farms when wind direction changes.

[0066] Regarding anomaly detection performance, two single-source detection methods were compared with the composite detection method proposed in this patent: (1) Cluster analysis method, based on K-means clustering, identifies data that deviates far from the cluster center as anomalies. (2) Statistical outlier detection method, based on the 3σ principle to identify outliers. The detection results are shown in Table 2. The fixed threshold method missed some minor power output anomalies at night in photovoltaic power plants, and the statistical outlier detection method failed to identify some logical consistency anomalies. The method proposed in this patent achieved a recall rate of 91.1% and an accuracy rate of 82.8%, demonstrating a more comprehensive anomaly detection capability.

[0067] Table 2 Comparison of Anomaly Data Identification Performance

[0068] Note: Recall refers to the percentage of truly anomalous data that is identified, while precision refers to the percentage of data identified as anomalous using this method that is actually anomalous.

[0069] Regarding data repair methods, two commonly used methods were selected for comparison with the method proposed in this patent: (1) linear interpolation method, which uses adjacent normal data for linear interpolation. (2) time series method, which uses a Long Short-Term Memory (LSTM) network model for single-site prediction and repair. The repair results are shown in Table 3. The analysis shows that the linear interpolation method performs the worst, and although the time series method can capture the time trend, its repair effect is limited when the entire site is affected by weather. The method of this patent dynamically learns the spatial relationship between sites through a graph attention mechanism, and still maintains good repair accuracy in complex scenarios such as wind direction changes.

[0070] Table 3 Comparison of Abnormal Data Repair Performance

[0071] Therefore, the multi-source verification method for new energy application data based on the GAT-CKDE algorithm proposed in this invention provides high-precision new energy output prediction data for the power grid dispatch center by constructing a verification system that integrates intelligent algorithms. This enables dispatch decision-making to shift from passively responding to deviations to actively managing uncertainty, significantly reducing the system's reserve and balancing costs and improving the effectiveness of dispatch tools. Simultaneously, the method proposed in this patent overcomes the problems of low efficiency and high human resource consumption of manual correction through an automated verification mechanism, helping to reduce the labor costs of verifying new energy application data. Ultimately, while ensuring power grid security, it achieves the optimization of total system operating costs and a dynamic balance between system operating efficiency and security.

[0072] Exemplary apparatus Figure 3 This is a schematic diagram of the structure of a multi-source verification device for new energy application data based on the GAT-CKDE algorithm, provided in an exemplary embodiment of the present invention. Figure 3 As shown, the device 300 includes: The identification module 310 is used to identify abnormal and missing data in the original new energy application data through a pre-constructed composite anomaly judgment function, and output marked abnormal data, wherein the abnormal data includes abnormal data and missing data; Repair module 320 is used to repair abnormal data using a pre-built graph attention network structure to obtain repaired new energy prediction data; The calculation module 330 is used to calculate the risk envelope of the repaired new energy prediction data using a pre-built risk envelope generation method based on conditional kernel density estimation of new energy prediction data.

[0073] Exemplary electronic device Figure 4 This is the structure of an electronic device provided in an exemplary embodiment of the present invention. For example... Figure 4 As shown, the electronic device 40 includes one or more processors 41 and a memory 42.

[0074] The processor 41 may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

[0075] The memory 42 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 41 may execute the program instructions to implement the methods of the software programs of the various embodiments of the present invention described above, and / or other desired functions. In one example, the electronic device may also include an input device 43 and an output device 44, these components being interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0076] In addition, the input device 43 may also include, for example, a keyboard, a mouse, etc.

[0077] The output device 44 can output various information to the outside. The output device 44 may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0078] Of course, for the sake of simplicity, Figure 4 Only some of the components of this electronic device relevant to the present invention are shown, omitting components such as buses, input / output interfaces, etc. In addition, the electronic device may include any other suitable components depending on the specific application.

[0079] Exemplary computer program product and computer readable storage medium In addition to the methods and apparatus described above, embodiments of the present invention may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform the steps in the methods according to various embodiments of the present invention described in the "Exemplary Methods" section above.

[0080] The computer program product can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of the present invention. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0081] Furthermore, embodiments of the present invention may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps of the methods according to various embodiments of the present invention described in the "Exemplary Methods" section above.

[0082] The computer-readable storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0083] The basic principles of the present invention have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in the present invention are merely examples and not limitations, and should not be considered as essential features of each embodiment of the present invention. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the present invention to the necessity of employing the aforementioned specific details.

[0084] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For system embodiments, since they largely correspond to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0085] The block diagrams of devices, systems, devices, and systems involved in this invention are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, systems, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0086] The methods and systems of the present invention may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the methods is for illustrative purposes only, and the steps of the methods of the present invention are not limited to the order specifically described above unless otherwise specifically stated. Furthermore, in some embodiments, the present invention may also be implemented as a program recorded on a recording medium, the program comprising machine-readable instructions for implementing the methods according to the present invention. Thus, the present invention also covers recording media storing programs for performing the methods according to the present invention.

[0087] It should also be noted that in the systems, apparatus, and methods of the present invention, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered equivalents of the present invention. The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the invention. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of the invention. Therefore, the invention is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0088] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the invention to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

Claims

1. A multi-source verification method for new energy application data based on the GAT-CKDE algorithm, characterized in that, include: The original new energy application data is identified as abnormal and missing by a pre-constructed composite anomaly detection function, and the abnormal data is marked as abnormal data and missing data. The abnormal data is repaired using a pre-constructed graph attention network structure to obtain repaired new energy prediction data; The risk envelope of the repaired new energy prediction data is calculated using a pre-constructed risk envelope generation method based on conditional kernel density estimation for new energy prediction data.

2. The method according to claim 1, characterized in that, The construction steps of the composite anomaly detection function include: Construct regular anomaly detection functions, logical consistency detection functions, and missing item detection functions; Based on the aforementioned regular anomaly determination function, logical consistency determination function, and missing value determination function, the composite anomaly determination function is constructed.

3. The method according to claim 1, characterized in that, The steps for constructing the regular anomaly detection function include: Construct functions for determining limits, mutations, and negative values; Based on the aforementioned over-limit determination function, mutation determination function, and negative value determination function, the regular anomaly determination function is constructed. The expression for the over-limit determination function is as follows: In the formula, d i,t For the data submitted by new energy power station i at time t, C i The sum of the nameplate values ​​of the equipment working at the station, where k1 is a configurable parameter set according to different regulatory rules; The expression for the mutation determination function is: In the formula, This is the power output dataset under similar weather conditions from time (t-T1) to time t; k2 is a configurable parameter set according to different regulatory rules; The expression for the negative value determination function is: The expression for the regular anomaly determination function is: When F r (d i,t When the output value of F is 1, the data point is marked as a regularity anomaly; when F... r (d i,t When the output value of ) is 0, the data point is considered to be normal data with regularity.

4. The method according to claim 3, characterized in that, The expression for the logical consistency determination function is: In the formula, θ i,t Let ε be the solar altitude angle at time t where the new energy power station i is located, and let c be a very small positive threshold set to avoid sensor noise or minor measurement errors. i For new energy type indication sign, c i =1 indicates that new energy source i is a photovoltaic power station, and c i =0 indicates that new energy source i is a wind farm. For process-level monitoring of SCADA data, For settlement-level measurement TMR data, α is the maximum permissible relative deviation threshold; when F l (d i,t When the value of F is 1, the data point is marked as having a logical consistency anomaly; when F... l (d i,t When the value of ) is 0, the data point is logically consistent and normal. The expression for the missing value determination function is F. m (d i,t If data is missing, record F. m (d i,t ) = 1, otherwise F m (d i,t ) = 0; The expression for the composite anomaly determination function is:

5. The method according to claim 1, characterized in that, The abnormal data is repaired using a pre-constructed graph attention network structure to obtain repaired new energy prediction data, including: All new energy power stations within the dispatch area are abstracted into a graph structure, resulting in a graph network structure; Calculate the initial edge weights based on the graph network structure, and construct the original adjacency matrix based on the initial edge weights; The original adjacency matrix is ​​subjected to self-looping and symmetric normalization to obtain a normalized adjacency matrix; Scan the feature vectors of all nodes. If the data quality mask bit in the feature vector is set to an abnormal state, then mark the node as the target node for repair. Centered on the target node to be repaired, a local subgraph is extracted, which includes the target node to be repaired and all its neighboring nodes; The local subgraph is input into a graph attention network model to dynamically calculate the attention weight of each neighbor node to the target node to be repaired; Based on the attention weight, the feature information of the neighboring nodes is weighted and aggregated to infer the missing or abnormal data of the target node to be repaired, thereby obtaining the new energy prediction data for repair.

6. The mechanism according to claim 1, characterized in that, The risk envelope of the repaired new energy prediction data is calculated using a pre-constructed risk envelope generation method based on conditional kernel density estimation, including: Based on historical actual output values ​​and historical predicted values, the prediction error at each historical time point is calculated to form a dataset containing historical predicted values ​​and historical prediction errors. For the current predicted value in the new energy prediction data to be repaired, by setting the window width, all historical prediction errors that meet the preset conditions are filtered out from the dataset to form a set of similar historical errors; Conditional kernel density estimation is applied to the error data in the similar historical error set to estimate the error probability density function under the current prediction conditions; The cumulative distribution function is calculated based on the conditional error probability density function, and the error quantile corresponding to the target confidence level is determined. Combined with the current prediction value, the risk envelope of the repaired new energy prediction data is generated.

7. A multi-source verification device for new energy application data based on the GAT-CKDE algorithm, characterized in that, include: The identification module is used to identify abnormal and missing data in the original new energy application data through a pre-constructed composite anomaly judgment function, and output marked abnormal data, wherein the abnormal data includes abnormal data and missing data; The repair module is used to repair the abnormal data using a pre-built graph attention network structure to obtain repaired new energy prediction data. The calculation module is used to calculate the risk envelope of the repaired new energy prediction data using a pre-built risk envelope generation method based on conditional kernel density estimation for new energy prediction data.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program for performing the method described in any one of claims 1-6.

9. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the method described in any one of claims 1-6.

Images