A power generation enterprise clearing capacity evaluation method and system based on multi-index fusion

Abstract

A power generation enterprise clearing capacity evaluation method and system based on multi-index fusion, which comprehensively calculates multi-dimensional indexes such as power generation capacity, stability, dispatching response and environmental adaptability; the above calculation results are used as nodes to build a graph structure, and the edge direction is defined; for each time step, the information of the current node and the neighbor nodes is aggregated to generate the representation vector of the current node; the edge weight is dynamically adjusted, and the single-step comprehensive graph representation of each time step is output; the single-step comprehensive graph representation is modeled into a multi-step comprehensive graph representation sequence in time sequence, and input into a graph neural network prediction module to obtain the new energy power generation enterprise clearing capacity score output at the future time step; the clearing capacity score sequence is judged for risk time steps, and the risk causes are analyzed; the operation scheme is adjusted, and the dispatching instruction of the future time step is output. The present application can realize dynamic evaluation and forward-looking dispatching optimization of new energy power generation enterprise clearing capacity, and improve the timeliness of risk identification and the pertinence of dispatching decision.

Description

Technical Field

[0001] This invention belongs to the field of power clearing capacity assessment technology, and in particular relates to a method and system for assessing the clearing capacity of power generation enterprises based on multi-indicator fusion. Background Technology

[0002] With the deepening of the global energy structure transformation, the installed capacity and penetration rate of new energy power generation such as wind power and photovoltaics in the power system have continued to grow rapidly. The participation and importance of new energy power generation companies in the electricity market are becoming increasingly prominent. In an electricity trading system centered on market-based clearing, whether new energy companies can successfully "clear" their power generation not only affects their own economic benefits and market competitiveness, but also directly impacts the supply-demand balance and safe and stable operation of the entire power grid. Therefore, accurately and dynamically assessing the "clearing capability" of new energy power generation companies has become a key technical issue for ensuring the stability of the new power system and promoting the consumption of new energy.

[0003] Currently, research and practice regarding electricity market clearing mainly focus on the system side and market mechanism level. For example, existing technologies, such as the patent document with publication number CN119918311B, focus on constructing a day-ahead market clearing optimization model that includes various aggregated resources (such as energy storage and interruptible loads) and renewable energy. Its core objective is to minimize the overall system operating cost and reduce the overall market risk caused by the volatility of new energy sources while meeting complex physical and operational constraints. The patent document with publication number CN120833060A focuses on the identification and assessment of market price risk in scenarios with a high proportion of new energy sources. It quantifies the volatility risk of market clearing prices by constructing a multi-dimensional indicator system and using data mining and comprehensive evaluation methods.

[0004] However, the aforementioned existing technologies share a common limitation: they primarily assess "market acceptance of new energy" or "market price volatility risk" from a macro perspective of grid dispatching agencies or market operators, lacking a dynamic and granular assessment of the intrinsic capabilities of "new energy power generation enterprises"—the micro-market entities themselves—to "meet market acceptance conditions." Specifically, existing technologies struggle to address the following pain points:

[0005] The assessment dimensions are too limited: traditional methods either rely on a single dimension such as historical clearing results or electricity price forecasts, or only focus on macro market models, failing to integrate and analyze the internal operating status (such as equipment health, power generation stability, and command response capability) and external environmental adaptability (such as forecast accuracy and anti-disturbance capability) that affect the company's clearing ability from multiple dimensions.

[0006] Lack of dynamism and foresight: Existing assessments are mostly static or post-hoc analyses, unable to integrate multi-source data in real time, dynamically capture the temporal evolution of enterprise operating status, and even less able to make probabilistic predictions and risk warnings about clearing capacity within future time windows.

[0007] The disconnect between "assessment" and "control": Existing methods usually stop at risk identification or capability scoring, failing to establish a closed loop from "capability assessment" to "risk source tracing" and then to "proactive control". They cannot provide enterprises with specific suggestions for operational optimization and scheduling instruction adjustment, and the assessment results are difficult to directly transform into scheduling decision-making basis for improving market competitiveness.

[0008] Therefore, how to design a method that can deeply integrate the internal operating status and external environmental conditions of new energy power generation enterprises to achieve dynamic, forward-looking, and closed-loop evaluation and optimization of their market clearing capabilities has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this application provides a method and system for assessing the clearing capacity of power generation enterprises based on multi-indicator fusion. By constructing a multi-dimensional indicator graph structure based on power generation equipment status, dispatch response, and meteorological conditions, and utilizing a time-series graph neural network to predict future clearing capacity, dynamic assessment and forward-looking dispatch optimization of the clearing capacity of new energy power generation enterprises are achieved.

[0010] The present invention adopts the following technical solution.

[0011] This invention provides a method for assessing the clearing-out capacity of power generation enterprises based on multi-indicator fusion, comprising:

[0012] Step S1: Obtain equipment operation data, dispatch response data, and environmental operation parameters of new energy power generation enterprises, and calculate multi-dimensional evaluation indicators that characterize the clearing-out ability of new energy power generation enterprises;

[0013] Step S2: Construct a graph structure using multidimensional evaluation metrics as nodes and determine edge directions. Update the edge weights for each time step to generate a single-step comprehensive graph representation for each time step.

[0014] Step S3: The single-step integrated graph representation is converted into a multi-step integrated graph representation sequence according to the time sequence, and the sequence is input into the graph neural network prediction module to obtain the probability value of meeting the grid dispatch acceptance conditions in the future time step, and the probability value is used as the clearing capability score.

[0015] Step S4: Identify the risk time step based on the clearing capability score, analyze the cause of the risk by combining the index values ​​of the previous time step, generate the equipment operation adjustment plan, and output the scheduling instructions for the future time step.

[0016] More preferably, the multi-dimensional evaluation indicators characterizing the clearing-out capability of new energy power generation enterprises in step S1 include:

[0017] The power generation capacity is calculated based on the rated power model of the power generation equipment, the equipment health factor, and the environmental correction coefficient. The equipment health factor is calculated by weighting the reliability value, the performance surplus coefficient, and the set positive adjustment term.

[0018] The power generation stability index is calculated based on the real-time power output sequence of the power generation equipment, the fault impact factor, and the capacity difference.

[0019] The execution scheduling capability is calculated based on the scheduling response delay, stabilization time, output error, and instruction type after the scheduling instruction is issued.

[0020] The power generation forecast accuracy is calculated based on the mean absolute error and root mean square error between the power generation forecast data and the actual data, after normalization and weighting.

[0021] The performance adaptability index is calculated based on the performance loss time and recovery time under disturbance conditions, after normalization and weighting.

[0022] More preferably, in step S2, the multidimensional evaluation index is used as a node to construct a graph structure and determine the edge directions, and the edge weights at each time step are updated, specifically as follows:

[0023] Multidimensional evaluation indicators are used as nodes to construct a graph structure, and the edge directions are determined based on the physical, scheduling, and environmental impact relationships among the multidimensional evaluation indicators.

[0024] For two nodes with an edge, the node correlation is calculated using a sliding window method based on the historical index values ​​of the nodes. The edge weights of the current time step are dynamically adjusted by weighting the edge weights of the previous time step with the correlation by a smoothing factor.

[0025] More preferably, the generation of the single-step synthesis diagram representation for each time step in step S2 is specifically as follows:

[0026] For each time step, the information of the current node and its neighboring nodes is aggregated, and combined with the node features of historical time steps, to generate the representation vector of the current node at the current time step;

[0027] At each time step, a single-step comprehensive graph representation is generated by aggregating the results from the graph convolutional network and dynamically adjusting the edge weights. The single-step synthetic graph representation includes node representation vectors and the updated edge weight matrix.

[0028] More preferably, in step S3, the multi-step synthetic graph representation sequence constructed from the single-step synthetic graph representation is input into the graph neural network prediction module, specifically as follows:

[0029] The single-step synthesis graph representations output from multiple time steps are arranged in chronological order to form a multi-step synthesis graph representation sequence;

[0030] The multi-step integrated graph representation sequence is input into the temporal coding layer to extract evolutionary features across time steps and output the hidden state of each node in the time step sequence.

[0031] Global pooling is performed on the hidden states of all nodes to obtain a global graph representation.

[0032] More preferably, in step S3, the probability value of satisfying the grid dispatch acceptance condition in the future time step is obtained, and the probability value is used as the clearing capability score, specifically as follows:

[0033] The global graph representation is input into the fully connected layer, and the output is the probability value of the new energy power generation enterprise meeting the grid dispatch acceptance conditions in the future time step. The probability value is used as the clearing capability score.

[0034] More preferably, in step S4, the risk time step is identified based on the clearing capacity score, and the causes of risk are analyzed in conjunction with the indicator values ​​of the preceding time step, specifically as follows:

[0035] The continuously output clearing capacity score is compared with the set threshold. If the clearing capability score output at any given time is lower than the set threshold, then... The time step is marked as a risk time step;

[0036] Retrieve Time before The indicator values ​​at each time step are analyzed, and the trend of change of the indicator values ​​is analyzed using time series analysis to determine the reasons for the decline in clearing capacity.

[0037] More preferably, in step S4, generating the equipment operation adjustment plan and outputting scheduling instructions for future time steps specifically involves:

[0038] Based on the identified risk causes, and using a pre-established cause-and-response mapping rule base, candidate adjustment strategies are matched for different risk causes.

[0039] Taking into account the current status of power generation equipment, meteorological conditions, and grid operation requirements, the final adjustment strategy is selected from the candidate adjustment strategies.

[0040] Based on the final adjustment strategy, a device operation adjustment plan is generated, and based on the preset mapping relationship, the device operation adjustment plan is converted into scheduling instructions.

[0041] This invention also provides a power generation enterprise clearing capacity assessment system based on multi-indicator fusion, the system comprising:

[0042] Multi-indicator calculation module: This module is responsible for calculating the power generation capacity and power generation stability indicators based on the operating status of the power generation equipment of new energy power generation enterprises; calculating the execution dispatch capability based on the response to power dispatch instructions; and calculating the power generation prediction accuracy and operation adaptability indicators based on the meteorological conditions and power system operating parameters of the region, so as to comprehensively evaluate the adaptability of the enterprise's operating status to the external environment.

[0043] Single-step graph modeling module: This module uses multiple computation results as nodes to construct a graph structure, and defines the direction of edges based on the physical, scheduling and environmental influence relationships between nodes. Within each time step, the information of the node and its neighboring nodes is aggregated through a graph convolutional network (GCN) to generate the node's representation vector, and the edge weights are dynamically adjusted based on the correlation of historical indicators, thereby outputting a single-step comprehensive graph representation for each time step, ensuring the effective transmission and modeling of time series information.

[0044] Clearing Capacity Assessment Module: This module models the single-step integrated graph representation into a multi-step integrated graph representation sequence based on the time series. The sequence is then input into the graph neural network prediction module integrated within the module. Through nonlinear mapping and aggregation operations, the probability value of the new energy power generation enterprise meeting the grid dispatch acceptance conditions in the future time step is obtained. This probability value serves as the clearing capacity score output of the new energy power generation enterprise, providing a scientific basis for grid dispatch.

[0045] Feedback Adjustment Module: This module performs risk time step judgment on the continuously output clearing capacity score sequence, compares the score at a given moment with a preset risk threshold, and marks that moment as a risk time step if the score is lower than the threshold; retrieves relevant indicator values ​​from previous time steps to analyze the causes of the risk; and generates equipment operation adjustment plans for new energy power generation enterprises based on the analysis results, outputs specific dispatch instructions, and ensures the stability of grid operation and the optimal dispatch strategy for new energy power generation enterprises.

[0046] More preferably, the present invention also provides a computer-readable storage medium storing a computer program thereon, characterized in that, when the computer program is executed by a processor, the steps of the above method are performed.

[0047] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0048] 1. This invention represents a fundamental shift from single static assessment to multi-dimensional dynamic perception. It abandons the traditional static assessment model that relies on a single historical clearing result or electricity price forecast. By uniquely integrating five core dimensions—equipment health factors, power generation stability, dispatch command response, prediction accuracy, and operational adaptability—it constructs a dynamic indicator system covering the entire "equipment-environment-dispatch" chain. This system not only quantifies the physical limits (power generation capacity) and controllability (dispatch execution capability) of new energy power generation in real time, but also, for the first time, incorporates the prediction uncertainty of external meteorological conditions (power generation prediction accuracy) and the ability to withstand disturbances (operational adaptability) into the assessment scope. This deep integration of multi-source heterogeneous data enables the assessment model to comprehensively and faithfully depict the true state and potential of new energy enterprises in complex operating environments, providing an unprecedented data foundation for subsequent accurate prediction and decision-making, and solving the fundamental problem of existing technologies having a single assessment dimension and failing to reflect dynamic risks.

[0049] 2. This invention constructs a time-series graph neural network model to characterize complex relationships, achieving probabilistic prediction of clearing capacity. This invention creatively constructs multi-dimensional indicators as a dynamic graph structure and utilizes a time-series graph neural network for modeling and prediction. The key technical aspects are: defining edges between nodes based on physical, scheduling, and environmental influence relationships, explicitly modeling the complex coupling mechanism between indicators in a computable manner for the first time; introducing a dynamic edge weight adjustment mechanism based on historical correlation, enabling the model to adaptively capture real-time changes in the influence between indicators; and aggregating multi-step historical graph sequence information through a time-series coding layer to deeply mine state evolution patterns. Ultimately, the model outputs the probability value (clearing capacity score) of "meeting the grid dispatch acceptance conditions" within a future time period. This prediction method not only yields more accurate results but also quantifies uncertainty in probabilistic form, providing dispatching agencies with a decision-making basis that combines foresight and risk confidence.

[0050] 3. This invention empowers proactive scheduling and risk prevention through an intelligent closed loop of "assessment-early warning-source tracing-regulation." This invention transcends the traditional "assessment-display" paradigm, constructing a complete intelligent decision-making closed loop. When the model predicts that the clearing capacity is below the risk threshold, the system automatically marks the "risk time step" and immediately retrieves preceding multi-dimensional indicator data for root cause analysis. Based on the clear root cause, the system combines real-time equipment status, meteorological conditions, and grid demand, and automatically generates specific equipment adjustment plans and parameterized scheduling instructions through a predefined "cause-countermeasure" rule base or optimization algorithm. This transforms scheduling decisions from passive response to proactive intervention, enabling early risk mitigation, optimization of power generation plans, and ultimately significantly improving the grid integration and absorption capacity of new energy sources, grid operation safety, and market operation efficiency, achieving the ultimate transformation of assessment value. Attached Figure Description

[0051] Figure 1 This is a flowchart of a method for assessing the clearing capacity of power generation enterprises based on multi-indicator fusion, according to the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0053] like Figure 1 As shown, this invention proposes a method for assessing the clearing capacity of power generation enterprises based on multi-indicator fusion, specifically including the following steps:

[0054] S1 calculates the power generation capacity and power generation stability index based on the operating status of power generation equipment in new energy power generation enterprises, calculates the execution dispatch capability based on the response to power dispatch instructions, and calculates the power generation prediction accuracy and operation adaptability index based on the meteorological conditions and power system operating parameters of the region.

[0055] In S1, the calculation process for the power generation capacity includes:

[0056] At the current time step, a rated power model is established for each power generation device. The calculation result of the rated power model is multiplied by the equipment health factor of the power generation device to obtain the basic power generation capacity of each power generation device.

[0057] An environmental correction factor is set and multiplied by the basic power generation capacity to obtain the power generation capacity of each power generation device; wherein, the environmental correction factor is obtained by empirical formula or regression of historical data.

[0058] The power generation capacity of all power generation equipment in the same new energy power generation enterprise is summed to obtain the power generation capacity of the new energy power generation enterprise at the current time step.

[0059] The method for calculating the health factors of the equipment includes:

[0060] The reliability value of the power generation equipment The calculation formula is: ,in The number of failures within a time step. For reference standard failure count;

[0061] The performance surplus factor of the power generation equipment The calculation formula is: ,in This represents the cumulative power attenuation rate.

[0062] The equipment health factor is obtained by weighting the reliability value, the performance residual coefficient, and the set positive adjustment term; the weights are set based on experience.

[0063] In S1, the calculation process of the power generation stability index includes:

[0064] At the current time step, calculate the coefficient of variation for each generator based on its real-time power output sequence.

[0065] By introducing a fault impact factor to correct the coefficient of variation, the power generation stability index of each device is obtained.

[0066] The power generation stability index of all power generation equipment in the same new energy power generation enterprise is weighted and summed according to the capacity difference. After normalizing the summation result, the power generation stability index of the new energy power generation enterprise at the current time step is obtained.

[0067] The steps for introducing a fault impact factor to correct the coefficient of variation and obtaining the power generation stability index for each power generation device include:

[0068] Calculate the total duration of the power generation equipment's failure or downtime within the time window. With time window Perform a ratio; the fault impact factor The calculation formula is: ,in For the device in the time window Total duration of internal faults or downtime.

[0069] Each power generation unit Power generation stability index The calculation formula is: ,in is the coefficient of variation of the real-time power output sequence.

[0070] In step S1, the method for calculating the execution scheduling capability is as follows:

[0071] Obtain all dispatch instructions and instruction types issued within the current time step from the power dispatch system, and simultaneously obtain the power output curve of the power generation equipment in the current time step;

[0072] Calculate the single-time scheduling response delay, stabilization time, and output error after each instruction is issued, and then weight them to obtain a single instruction execution capability score;

[0073] The execution capability scores of all instructions within the current time step are weighted and averaged according to the instruction type to obtain the execution and scheduling capability of the new energy power generation enterprise.

[0074] In S1, the specific calculation method for the power generation prediction accuracy is as follows:

[0075] Acquire power generation forecast data and actual data from new energy power generation enterprises for a future set time period; the power generation forecast data is obtained based on the forecasting algorithm actually deployed in the new energy power generation enterprises; calculate the mean absolute error and root mean square error between the power generation forecast data and the actual data;

[0076] The normalized mean absolute error and root mean square error are weighted and summed; the power generation prediction accuracy is... The calculation formula is: ,in and These are the normalized mean absolute error and root mean square error, respectively. and The corresponding weights.

[0077] In S1, the specific calculation method for the operational adaptability index is as follows:

[0078] Obtain meteorological data and power system operating parameters of the area where the new energy power generation enterprise is located, determine the set of disturbance scenarios, set an assessment period for each disturbance scenario, and calculate the performance loss time and recovery time within the assessment period;

[0079] Specifically, the moment when the disturbance begins to have an observable negative impact on the system is taken as the starting moment of the performance loss, and the moment when the performance index drops to its lowest point or enters a stable low value state is taken as the ending moment of the performance loss, thus obtaining the performance loss time of the disturbance.

[0080] The recovery time for the disturbance is calculated from the moment the performance loss ends until the system performance index recovers to a stable level.

[0081] The performance loss time and recovery time for each disturbance are normalized to the 0-1 interval, and a weighted average is performed to obtain the total performance loss and average recovery time.

[0082] The operational adaptability indicators The calculation formula is: ,in The total performance loss after normalization. This represents the normalized mean recovery time.

[0083] S2 uses the power generation capacity, power generation stability index, execution scheduling capability, power generation prediction accuracy and operation adaptability index as nodes to construct the graph structure and defines the edge directions; for each time step, it aggregates the information of the current node and its neighboring nodes, and uses the node features of historical time steps to generate the representation vector of the current node in the current time step; it dynamically adjusts the edge weights according to the correlation of historical indicators between nodes, and outputs the single-step comprehensive graph representation of each time step.

[0084] In S2, the method of dynamically adjusting edge weights based on the correlation of historical indicators between nodes and outputting a single-step comprehensive graph representation for each time step includes:

[0085] For two nodes with an edge, the correlation is calculated using a sliding window based on the historical node index values ​​of the nodes, and the edge weights at the current time step are dynamically adjusted.

[0086] The dynamic adjustment method for edge weights is as follows: [The rest of the text appears to be a list of steps and requires further context for accurate translation.] and At time step The edge weights are multiplied by the set smoothing factor to obtain the first product result; the difference (1 minus the smoothing factor) is multiplied by the correlation between historical node index values ​​to obtain the second product result; the first and second product results are added together to obtain the node... and At time step Edge weights;

[0087] At each time step, a single-step comprehensive graph representation is generated by aggregating the results from the graph convolutional network and dynamically adjusting the edge weights. This includes node representation vectors and the updated edge weight matrix.

[0088] S3, the single-step integrated graph representation is modeled into a multi-step integrated graph representation sequence according to the time sequence, and input into the graph neural network prediction module with a probability output layer for nonlinear mapping and aggregation operations to obtain the probability value of the new energy power generation enterprise meeting the grid dispatch acceptance conditions in the future time step; the probability value is used as the clearing capability score output of the new energy power generation enterprise;

[0089] In S3, the single-step synthesis graph representations output from multiple time steps are arranged in temporal order to form a multi-step synthesis graph representation sequence. Each single-step synthetic graph representation includes node feature vectors and a dynamic edge weight matrix.

[0090] Will Input a temporal coding layer to extract evolutionary features across time steps, and output the hidden state of each node in the time step sequence; perform global pooling on the hidden states of all nodes to obtain a global graph representation;

[0091] The global graph representation is input into a fully connected layer with nonlinear activation, and the output is the probability that a new energy power generation company will meet the grid dispatch acceptance conditions in a future time step.

[0092] S4 performs risk time step judgment on the clearing capacity scoring sequence, analyzes the risk causes by combining the index values ​​of the previous time step, generates equipment operation adjustment plans for new energy power generation enterprises based on the analysis results, and outputs scheduling instructions for future time steps.

[0093] In S4, the method for determining the risk time step of the clearing capacity scoring sequence and analyzing the causes of risk in conjunction with the indicator values ​​of the preceding time step is as follows:

[0094] The continuously output clearing capacity score is compared with the set threshold. If the clearing capability score output at any given time is lower than the set threshold, then... The time step is marked as a risk time step;

[0095] Retrieve Time before The system analyzes the risk causes based on the indicator values ​​at each time step; it then outputs equipment operation adjustment plans for new energy power generation enterprises based on the analyzed risk causes, and transforms these plans into dispatch instructions.

[0096] This invention also provides a power generation enterprise clearing capacity assessment system based on multi-indicator fusion, employing the specific implementation methods disclosed in the above embodiments. The system includes:

[0097] Multi-indicator calculation module: This module is responsible for calculating the power generation capacity and power generation stability indicators based on the operating status of the power generation equipment of new energy power generation enterprises; calculating the execution dispatch capability based on the response to power dispatch instructions; and calculating the power generation prediction accuracy and operation adaptability indicators based on the meteorological conditions and power system operating parameters of the region, so as to comprehensively evaluate the adaptability of the enterprise's operating status to the external environment.

[0098] Single-step graph modeling module: This module uses multiple computation results as nodes to construct a graph structure, and defines the direction of edges based on the physical, scheduling and environmental influence relationships between nodes. Within each time step, the information of the node and its neighboring nodes is aggregated through a graph convolutional network (GCN) to generate the node's representation vector, and the edge weights are dynamically adjusted based on the correlation of historical indicators, thereby outputting a single-step comprehensive graph representation for each time step, ensuring the effective transmission and modeling of time series information.

[0099] Clearing Capacity Assessment Module: This module models the single-step integrated graph representation into a multi-step integrated graph representation sequence based on the time series. The sequence is then input into the graph neural network prediction module integrated within the module. Through nonlinear mapping and aggregation operations, the probability value of the new energy power generation enterprise meeting the grid dispatch acceptance conditions in the future time step is obtained. This probability value serves as the clearing capacity score output of the new energy power generation enterprise, providing a scientific basis for grid dispatch.

[0100] Feedback Adjustment Module: This module performs risk time step judgment on the continuously output clearing capacity score sequence, compares the score at a given moment with a preset risk threshold, and marks that moment as a risk time step if the score is lower than the threshold; retrieves relevant indicator values ​​from previous time steps to analyze the causes of the risk; and generates equipment operation adjustment plans for new energy power generation enterprises based on the analysis results, outputs specific dispatch instructions, and ensures the stability of grid operation and the optimal dispatch strategy for new energy power generation enterprises.

[0101] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0102] Example 1

[0103] As an embodiment of this application, a specific implementation method for assessing the clearing capacity of power generation enterprises based on multi-indicator fusion is disclosed. For example... Figure 1 As shown, the method includes the following steps:

[0104] S1: For new energy power generation enterprises, the power generation capacity and power generation stability indicators are calculated based on the operating status of the power generation equipment, the execution of dispatching capacity is calculated based on the response to power dispatching instructions, and the power generation prediction accuracy and operation adaptability indicators are calculated based on the meteorological conditions of the region and the operating parameters of the power system.

[0105] S1.1: At the current time step For each power generation device in a new energy power generation enterprise, a rated power model is established based on the device type, capacity, and design parameters. Then, the model is corrected based on the health status of the power generation device to estimate the basic power generation capacity of a single device.

[0106] Assume the total number of power generation equipment in a new energy power generation company is Then for the first one The rated power model of the generator is constructed as follows: Optionally, if the equipment type is photovoltaic, the rated power model is a photovoltaic conversion model; if it is a wind turbine, the rated power model can be a wind energy conversion model.

[0107] Based on rated power model Power generation equipment The basic power generation capacity is estimated as follows ;in, The equipment health factor is calculated as follows:

[0108] Calculate reliability value based on failure frequency ; The number of failures within a time step; For reference, the number of failures can be calculated based on historical averages or equipment lifespan.

[0109] Calculate the performance residual factor based on runtime and decay rate. ; The cumulative output attenuation rate is an indicator that reflects the cumulative decrease in the actual maximum power generation capacity of a power generation device over time due to aging, wear, or efficiency decline. This parameter can be defined and calculated based on the type of power generation device (wind turbine, photovoltaic panel), available data (historical performance test records, manufacturer datasheets), and industry-standard evaluation criteria. It will not be elaborated on further here.

[0110] The health factor of the power generation equipment can be expressed as: ; and These are the weighting coefficients for the reliability value and the performance residual coefficient, respectively. This is a positive adjustment item, used to consider the effects of regular maintenance, preventative maintenance, and other practices on restoring and improving the health of the equipment. Among them, and Weight values ​​are directly assigned by those skilled in the art based on historical experience and an understanding of the criticality of different equipment types. Furthermore, a correlation model is established between maintenance activities and equipment performance recovery; for example, analyzing the improvement in equipment output or efficiency after each maintenance cycle, thereby assigning different weights to different types of maintenance activities. Value, therefore This conclusion is based on experimental performance.

[0111] Furthermore, considering the impact of the external environment on new energy power generation equipment, an environmental correction coefficient is introduced to correct the basic power generation capacity, thus obtaining the power generation capacity of a single power generation unit. , is represented as:

[0112] ;

[0113] The environmental correction coefficient can be calculated using empirical formulas or historical data regression, such as random forests, gradient boosting trees, or neural network models. The environmental factors include, but are not limited to, light intensity, wind speed, and temperature.

[0114] All within the same enterprise The sum of the power generation capacities of each power generation device represents the power generation capacity of the new energy power generation company at the current time step. .

[0115] S1.2: At the current time step, calculate the coefficient of variation for each device based on the real-time power output sequence. The coefficient of variation is corrected by introducing fault impact factors to obtain the power generation stability index.

[0116] As one implementation method of this embodiment, the fault impact factor is calculated based on the fault duration of the power generation equipment within a time step, and is expressed as follows: ; Indicates time window Total duration of internal equipment failures or downtime. Time window. The text inside represents the current time step. Compared with the previous time step The time difference between them.

[0117] The power generation stability index is expressed as ;in, Indicates power generation equipment At the current time step The power generation stability index.

[0118] The power generation stability index of all power generation equipment within the same enterprise is weighted and summed according to capacity differences, and then normalized. This summation yields the power generation stability index of the new energy power generation enterprise at the current time step, expressed as: The weighted summation based on capacity differences means calculating the proportion of a particular device in the company's total capacity as its weight.

[0119] S1.3: Obtain all dispatch instructions issued within the current time step from the power dispatch system, including instruction issuance time, target power value, dispatch duration, and instruction type (e.g., power increase, power decrease, power maintenance, etc.); synchronously, retrieve the power output curve of the generating equipment in the current time step.

[0120] The single-time scheduling response delay, stabilization time, and output error are calculated and weighted to obtain a single instruction execution capability score; the weights here can be set according to the emphasis in actual application.

[0121] The execution capability scores within the current time step are weighted and averaged according to instruction type to obtain the execution dispatch capability of new energy power generation enterprises. The weights for different instruction types can be determined using methods such as the analytic hierarchy process (AHP), regression fitting, or domain experience. In a preferred embodiment of this invention, basic weights are assigned to different instruction types based on the priority and safety importance of various instructions assessed by the power grid dispatching agency. For example, dispatching instructions can be divided into three categories: "emergency adjustment instructions," "planned power instructions," and "power maintenance instructions." "Emergency adjustment instructions" have the highest requirement for response speed and are assigned the highest weight (e.g., 0.5); "planned power instructions" have a relatively high requirement for process stability and are assigned a medium weight (e.g., 0.3); "power maintenance instructions" mainly assess steady-state accuracy and are assigned a basic weight (e.g., 0.2). This weight allocation scheme can be determined by expert scoring based on the industry-standard AHP, or obtained through regression fitting of historical dispatch assessment data; these are existing mature technologies and will not be elaborated further here. When weighting and averaging the execution capability scores, the calculation is performed according to the weight corresponding to the instruction type.

[0122] S1.4: Obtain meteorological data, power system operating parameters, and power generation forecast and actual data for the region where the new energy power generation enterprise is located; calculate the mean absolute error (MAE) and root mean square error (RMSE), and calculate the power generation forecast accuracy according to the following formula. :

[0123] ;

[0124] In the formula, and These are the normalized mean absolute error (MAE) and root mean square error (RMSE), respectively. and These are the weights for the normalized mean absolute error (MAE) and root mean square error (RMSE), respectively, assigned empirically based on error characteristics. If the evaluation focuses more on typical prediction bias levels, a higher weight can be given to MAE (e.g., ...). =0.7, =0.3); If the assessment is more vigilant and penalizes occasional large errors (such as severe prediction failures), then RMSE can be given a higher weight (e.g., =0.7, =0.3).

[0125] The power generation forecast data comes from the predicted power curve output by the short-term power forecasting system deployed at the new energy power generation enterprise. The short-term power forecasting system generates predicted values ​​of new energy power generation for specific future periods (such as the next 15 minutes, 1 hour, or 24 hours) based on numerical weather prediction (NWP) data, historical power data, equipment status, and other data.

[0126] S1.5: Based on the meteorological conditions and power system operating parameters of the region, determine the set of disturbance scenarios, and calculate the performance loss and recovery time during the disturbance for each scenario. Calculate the total performance loss and average recovery time for all disturbance scenarios, and combine them after normalization into the aforementioned operational adaptability index.

[0127] As one implementation method of this embodiment, based on historical meteorological records and power grid operation events in the area where the new energy power generation enterprise is located, several typical disturbance scenarios are identified. These include sudden strong winds, sudden reduction in sunlight, and line congestion. An assessment period is set for each scenario, including performance loss time and recovery time.

[0128] Specifically, within the assessment period, the moment when the disturbance begins to have an observable negative impact on the system (in this case, new energy power generation) is taken as the starting point of performance loss, and the moment when the performance index drops to its lowest point or enters a stable low value state is taken as the ending point of performance loss, thus obtaining the performance loss time. The moment when an observable negative impact occurs can be defined as follows: when the actual power is continuously lower than the predicted power or the dispatch command value exceeds a preset deviation threshold (e.g., 10%) and persists for more than a minimum time (e.g., 1 minute), performance loss is determined to begin; when the actual power reaches its lowest point and begins to show a stable upward trend, or when the power value recovers to the predicted power / dispatch command value minus a deviation threshold (e.g., still below 5%), performance loss is determined to end.

[0129] The recovery time is calculated from the moment the performance loss ends until the system performance index recovers to an acceptable stable level. The moment the system performance index recovers to an acceptable stable level can be defined as when the actual power recovers and remains stable within an acceptable deviation range (e.g., ±2%) of the predicted power / dispatch command value for more than a certain period of time (e.g., 5 minutes), at which point the recovery is considered complete.

[0130] As an alternative implementation, one or more power data sources, such as equipment status signals, grid electrical quantities, or dispatch control commands, can be combined to identify the start of disturbances and define the recovery status earlier and more accurately. Simultaneously, multiple power data sources can be used in combination for joint judgment. For example, a logical rule can be defined: the performance loss period is formally determined to have begun only when both "actual power is below a threshold" and ("equipment alarm signal triggered," "grid electrical quantity exceeds limit," or "specific dispatch command received") are simultaneously met. This improves the accuracy of the judgment and its anti-interference capability. Similar composite conditions can also be used to determine the recovery period.

[0131] In a further implementation, the performance loss time and recovery time for each disturbance scenario are normalized to the 0-1 interval, and a weighted average is performed to obtain the total performance loss and average recovery time. The weights can be set using the probability of occurrence or the severity of the disturbance. Specifically:

[0132] Based on probability of occurrence: the frequency of various disturbance events in historical data. For example, if "sudden decrease in sunlight" occurs 100 times and "line congestion" occurs 20 times in historical records, then the weight of "sudden decrease in sunlight" can be set to 100 / (100+20)≈0.83, and the weight of "line congestion" can be set to 0.17.

[0133] Based on severity: Define a quantitative standard for "severity," such as measuring "average power generation loss per event (MWh)." Assuming an average loss of 100 MWh per "sudden strong wind" and an average loss of 50 MWh per "sudden decrease in sunlight," the weight of "sudden strong wind" can be set to 100 / (100+50)≈0.67, and the weight of "sudden decrease in sunlight" can be set to 0.33.

[0134] Overall weight: Probability and severity can also be combined, for example: Overall weight = Probability weight × 0.4 + Severity weight × 0.6.

[0135] Based on the principle of "smaller loss, faster recovery, higher score", the normalized total performance loss and mean recovery time are combined to obtain the operational adaptability index. The comprehensive approach used here is expressed as follows:

[0136] ;

[0137] in, and These represent the total performance loss and the mean recovery time, respectively.

[0138] This invention abandons the traditional single-dimensional assessment method and constructs a multi-dimensional, fine-grained indicator system covering equipment status, dispatch response, weather forecasting, and anti-interference capabilities. This breaks through the limitations of the traditional assessment method where equipment, dispatch, and environmental factors are isolated from each other, and achieves a comprehensive and detailed characterization of the operating status of new energy power generation enterprises, providing a comprehensive and reliable underlying data foundation for subsequent accurate assessment and prediction.

[0139] S2: Multiple calculation results are used as nodes to construct a graph structure. Edge directions are defined based on the physical, scheduling, and environmental influence relationships between nodes. For each time step, information about the node and its neighbors is calculated. Edge weights are dynamically adjusted based on the correlation of historical indicators between nodes, and a single-step comprehensive graph representation for each time step is output. Specifically:

[0140] The five indicators obtained from S1—including power generation capacity, power generation stability, execution dispatch capability, power generation prediction accuracy, and operational adaptability—are used as nodes to construct a graph structure. The existence and direction of edges are defined based on the physical, dispatch, and environmental impact relationships between nodes.

[0141] Using the five indicators obtained from S1 (power generation capacity, power generation stability, execution dispatch capability, power generation prediction accuracy, and operational adaptability) as nodes, a directed graph is constructed. ,in For a set of nodes, Let this be the set of edges. The direction is from node i to node j, indicating that node i has a direct influence on node j.

[0142] The definition and direction determination of edges follow the following steps to standardize the rules:

[0143] 1. Based on physical influence relationships. For example, "power generation capacity" is the physical upper limit of "execution scheduling capacity", so a directed edge is established from the "power generation capacity" node to the "execution scheduling capacity" node.

[0144] 2. Based on scheduling influence relationships. The level of "execution scheduling capability" affects the "generation stability index" (e.g., frequent adjustments may lead to power fluctuations), and conversely, the "generation stability index" also affects the executability of scheduling instructions. Therefore, a bidirectional edge is established between these two nodes.

[0145] 3. Based on Environmental Impact Relationships. "Operational Adaptability Indicators" reflect the ability to resist and recover from external environmental disturbances (such as meteorological disturbances), directly affecting "Power Generation Capacity" and "Power Generation Stability Indicators." Therefore, two directed edges are established from the "Operational Adaptability Indicators" node to the "Power Generation Capacity" and "Power Generation Stability Indicators" nodes. Simultaneously, "Power Generation Forecast Accuracy" depends on the accuracy of environmental forecasts, and its level directly affects the degree to which dispatch decisions rely on "Operational Adaptability Indicators." Therefore, a directed edge is also established from the "Power Generation Forecast Accuracy" node to the "Operational Adaptability Indicators" node. Environmental conditions themselves are not treated as graph nodes but rather as external parameters for calculating these indicators; their impact is indirectly reflected in the graph structure through the temporal changes in indicator values.

[0146] In summary, for the five-node graph in this embodiment, its adjacency matrix A is defined as follows (rows represent the starting point, columns represent the ending point, 1 indicates the existence of a directed edge, and 0 indicates the absence of a direct edge):

[0147] Table 1 Examples of Five-Node Diagrams

[0148]

[0149] At the current time step For nodes Aggregate its neighboring nodes through a graph convolutional network. The node feature information, combined with the node At time step Node characteristics, generate nodes The representation vector at the current time step ;when Time step Node features A vector based on the results of index calculation; Represents a node The set of neighboring nodes;

[0150] In a further implementation, for two nodes with an edge, the correlation is calculated using a sliding window based on the historical node index values ​​of the nodes, and the edge weights at the current time step are dynamically adjusted. The correlation between the historical node index values ​​can be determined using the Pearson correlation coefficient or other techniques such as mutual information. The sliding window size can be set to 24 time steps; for example, when the time step is 15 minutes, the window covers historical data from the past 6 hours; when the time step is 1 hour, it covers historical data from the past 24 hours. The dynamic adjustment of the edge weights can be expressed by the following formula:

[0151] ;

[0152] In the formula, For nodes and At time step edge weights, For time steps edge weights, The correlation between historical node indicator values; As a smoothing factor, it controls the weight ratio between historical edge weights and relevance. In a preferred embodiment of the present invention, the smoothing factor... It can be set to 0.8 to balance the impact of historical weights and the latest correlations.

[0153] At each time step, a single-step comprehensive graph representation is generated by aggregating the results from the graph convolutional network and dynamically adjusting the edge weights. This includes node representation vectors and the updated edge weight matrix:

[0154] ;

[0155] In this embodiment, by defining the direction of edges and considering the relationships between physical, scheduling, and environmental influences, the complex relationships between indicators can be effectively modeled, providing a precise foundation for decision support and scheduling instruction generation. Dynamically adjusting edge weights enhances the model's sensitivity to historical data, thereby improving the accuracy of clearing capacity prediction. Graph convolutional networks, by integrating data from different time steps, generate a single-step comprehensive graph representation, capturing temporal features and providing robust input for multi-step prediction. This model can dynamically and quantitatively capture and characterize the complex coupling relationships and evolutionary patterns between indicators, significantly improving the model's ability to represent the dynamic characteristics of the system and providing a more scientific basis for prediction. Dynamically adjusted edge weights enhance the model's robustness and improve its adaptability under different environments and tasks.

[0156] S3: The single-step integrated graph representation is modeled into a multi-step integrated graph representation sequence according to the time sequence, and input into the graph neural network prediction module for nonlinear mapping and aggregation operations to obtain the probability value of the new energy power generation enterprise meeting the grid dispatch acceptance conditions in the future time step; the probability value is used as the clearing capability score of the new energy power generation enterprise.

[0157] The graph neural network prediction module consists of the following three layers in sequence:

[0158] (1) Temporal coding layer: used to extract evolutionary features across time steps. In this embodiment, a bidirectional gated recurrent unit (Bi-GRU) is used, and the input is a multi-step synthesized graph representation sequence. The output is the hidden state of each node in the time step sequence.

[0159] (2) Global pooling layer: Perform global pooling on the hidden states of all nodes to obtain a global graph representation. This embodiment uses weighted average pooling.

[0160] (3) Probabilistic Output Layer: Consists of a fully connected layer with a non-linear activation function (such as Sigmoid). It represents the global graph. The probability value is mapped to a value between [0,1], which is the clearing capability score.

[0161] Specifically, the single-step synthesis graph representations output from multiple time steps are arranged sequentially to form a multi-step synthesis graph representation sequence, in the form of: ;in, This represents the total number of time steps; each single-step composite graph representation includes a node representation vector and a dynamic edge weight matrix.

[0162] Will Input a temporal coding layer to extract evolutionary features across time steps, and output the hidden state of each node in the time step sequence;

[0163] In a further embodiment, the temporal coding layer employs a temporal feature extraction model, which may be a bidirectional gated recurrent unit (GRU), a long short-term memory network (LSTM), or other methods. The temporal feature extraction model extracts feature information between multiple single-step synthesized graph representations.

[0164] In a further implementation, the hidden states of all nodes are globally pooled, which can be done using weighted average pooling, to obtain a global graph representation. Representing the global graph The input is mapped to a fully connected layer with nonlinear activation, and the output is a sequence of probability values ​​for future time steps. This sequence of probability values ​​represents the probability prediction of whether new energy power generation companies can accept grid dispatch instructions within the future dispatch window, which is also the clearing capacity score sequence.

[0165] The core learning objective of the graph neural network prediction module is for new energy power generation enterprises to meet the grid dispatch acceptance conditions within the future time step. This objective is not directly input into the model as an explicit rule, but is implicit in the model training process. Specifically, during the model training phase, historical time-series data needs to be prepared, and its labels are generated based on whether the power generation enterprises actually met the grid dispatch requirements at that time. The model is trained using massive amounts of such "feature-label" pairs, thereby autonomously learning the complex mapping rules from complex multi-dimensional indicator states to the final comprehensive result of whether the conditions are "met". Therefore, the probability value output by the model (i.e., the clearing capability score) is essentially a quantitative prediction made by the model based on the current and historical operating states of the probability of the event of meeting the conditions in the future.

[0166] The criteria for accepting grid dispatch are the specific, quantifiable technical standards upon which the training labels are generated. These criteria can be defined according to the actual procedures of grid dispatch, and typically include, but are not limited to, the following categories:

[0167] Power tracking accuracy criteria: Within the dispatch time window, the error (such as mean absolute error, root mean square error) between the actual output of the power generation enterprise and the AGC command or power generation plan curve issued by the dispatching agency must be kept within the specified technical standards (for example, the error shall not exceed ±2% or ±5%).

[0168] Dynamic response performance criteria: The adjustment rate of the power generation enterprise in response to dispatch instructions must meet the minimum requirements (e.g., not less than 2% / minute of rated capacity), and the time from the issuance of the instruction to reaching the target and stabilizing (response delay and stabilization time) must be within the specified range.

[0169] Steady-state operation reliability criterion: After reaching the target output, the power generation company must be able to maintain stable operation for a specified minimum duration within the allowable steady-state error band, and its output should not experience any non-mandatory abnormal fluctuations or drops.

[0170] Adjustable capacity adequacy criterion: At any given time, the actual upward and downward reserve capacity that a power generation company can provide must be able to simultaneously meet the real-time reserve capacity requirements issued by the dispatching agency to ensure system safety and balance.

[0171] Multi-dimensional technical performance comprehensive criteria: Multiple key operating indicators of power generation enterprises (such as power generation capacity, power generation stability, prediction accuracy, etc.) must simultaneously meet a set of preset minimum threshold values, which reflects a more comprehensive "dispatchable" state.

[0172] Such criteria not only provide clear supervision signals for model training, but also make the final output clearing capability score have clear and intuitive engineering significance, directly representing the probability that power generation companies will be able to execute dispatch instructions well in the future and thus successfully achieve market clearing.

[0173] It should be noted that the implementation of this invention mainly revolves around the inference phase of the model. Before actual deployment, the above-mentioned graph neural network prediction module needs to be trained offline using a training dataset containing historical indicator data and their corresponding "whether the power grid dispatch acceptance conditions are met" labels. After training, the model is solidified and deployed in the online system. During actual operation, the system only loads the trained model parameters, executes the aforementioned inference process S2 to S4 based on real-time collected and calculated multi-dimensional evaluation indicators, outputs the clearing capability score, and triggers subsequent early warning and control actions. Online training updates of the model, such as incremental learning, can be considered as optional enhancements, but do not constitute a necessary step in the implementation of this invention.

[0174] This embodiment uses a single-step composite graph representation of multiple historical time steps for time series modeling, which can accurately capture the evolution characteristics in the time series, thereby improving the prediction accuracy of the clearing capacity of new energy power generation enterprises. Simultaneously, nonlinear mapping and aggregation operations integrate historical information from different time steps, providing strong decision support for grid dispatching.

[0175] This invention combines single-step graph representations into multi-step comprehensive graph representation sequences based on time series, and inputs these sequences into a time-series graph neural network for prediction. This method not only captures the spatial correlations between indicators (through graph structure) but also deeply explores their temporal evolution patterns. The final output, the "probability value of meeting the grid dispatch acceptance conditions" (i.e., clearing capacity score), enables forward-looking, probabilistic, and accurate prediction of the future clearing capacity of new energy power generation enterprises, providing direct and quantitative forward-looking basis for dispatch decisions.

[0176] S4: Perform risk time step judgment on the clearing capacity scoring sequence, analyze the risk causes by combining the index values ​​of the previous time step; generate equipment operation adjustment plan for new energy power generation enterprises based on the analysis results, and output scheduling instructions for future time steps.

[0177] The continuously outputting clearing capability score sequence is represented as follows: For each time step, a predicted value is given. In a typical implementation, the time step can be set to 15 minutes or 1 hour to match the common clearing cycle and dispatch instruction cycle in the electricity market; a risk threshold is set. The risk threshold can be set based on historical data, enterprise standards, or power grid dispatching requirements. In a preferred embodiment of the invention, the risk threshold can be set to 0.7 (or 70%). That is, when the clearing capability score (probability value) output by the model is lower than 0.7, that moment is marked as a risk time step, and an early warning and cause analysis process is triggered.

[0178] For each time step, its clearing capacity score is compared with a risk threshold. If it is lower than the risk threshold, the time step is marked as a "risk time step," indicating that the clearing capacity of new energy power generation enterprises cannot meet the grid dispatching needs at that time step.

[0179] For time steps marked as "risk time steps" retrieve its previous The system analyzes the generating capacity, generation stability indicators, dispatch execution capability, generation forecast accuracy, and operational adaptability indicators at each time step. Time-series analysis methods (including but not limited to sliding window statistical analysis, abrupt change detection algorithms, correlation / contribution analysis, and model-based feature importance analysis, or a combination of these methods) are used to analyze the changing trends of these indicators and determine the reasons for the decline in clearing capacity. The value can be set to 12, if the previous Within a certain time step, the power generation capacity decreased significantly, possibly due to health factors of the power generation equipment or changes in meteorological conditions.

[0180] In a further implementation, based on the risk factors, and taking into account the current status of the power generation equipment, meteorological conditions, and grid operation requirements, an equipment operation adjustment plan is generated and the equipment operation adjustment plan is transformed into a dispatch instruction.

[0181] Specifically, based on a pre-established "cause-countermeasure" mapping rule base, one or more candidate adjustment strategies are matched for different risk causes. Using methods including but not limited to decision tables or if-then rules, the adjustment strategies are transformed into specific, parameterized dispatch instructions that can be directly identified and executed by the power dispatching system or station monitoring system.

[0182] As a preferred embodiment of the present invention, the process of matching one or more candidate adjustment strategies for different risk causes based on a pre-established cause-countermeasure mapping rule base is to transform the qualitative or quantitative risk causes obtained from the analysis into specific and executable scheduling instructions according to a predefined cause-countermeasure mapping rule base. The specific implementation steps are as follows:

[0183] The rule base is established through inductive summarization based on power system operation procedures, equipment technical specifications, and the experience of dispatching experts. Its logical structure is usually expressed as an "IF-THEN" production rule or decision table. The antecedent (IF part) of the rule is the judgment condition for the risk cause (usually associated with the abnormal state of a specific assessment indicator, such as "continuous decline in generating capacity"), and the consequent (THEN part) is one or more corresponding candidate adjustment strategies (such as "starting standby units" or "increasing the active power setpoint").

[0184] Taking the typical risk factor of "the power generation capacity index being consistently low due to predicted wind speeds being lower than actual values" as an example, the mapping and instruction generation process is as follows:

[0185] Calculate the linear regression slope of the power generation capacity index for the first N time steps of the risk time step. If the slope is negative and the absolute value exceeds the threshold A (e.g., slope < -0.05 units / time step), it is determined to be a continuous decline.

[0186] Calculate the rolling correlation coefficient (e.g., Pearson correlation coefficient) between the two indicator sequences of generating capacity and generating forecast accuracy within the same time window. If the average value exceeds the positive correlation threshold B (e.g., correlation coefficient > 0.7), then a strong positive correlation is determined between the two.

[0187] Read the numerical weather prediction (NWP) correction data to confirm that the actual wind speed is consistently lower than the predicted value and the deviation exceeds the preset threshold C (e.g., average wind speed deviation > 15%).

[0188] Furthermore, rules that meet the following combined conditions are matched in the rule base:

[0189] IF (Power Generation Capacity_Trend == "Continued Decline");

[0190] AND (Power generation capacity and power generation forecast accuracy are strongly positively correlated).

[0191] AND(Wind speed prediction deviation status == "Actual speed remains consistently low");

[0192] THEN candidate strategy = {"Upgrade short-term power generation plan", "Request to discharge energy storage within the power station"};

[0193] Based on the current real-time demand of the power grid and the status of equipment, the system selects "upward revision of short-term power generation plan" from the candidate strategies. Subsequently, the strategy is converted into a parameterized scheduling instruction, such as: "Increase the baseline value of the short-term power generation plan curve for the next 60 minutes of [XX Wind Farm] by [the compensation value of the current prediction deviation, such as 8%]".

[0194] Through the above mechanism, an intelligent closed loop from "assessment-early warning" to "diagnosis-control" is realized, enabling the assessment results to directly drive operational optimization.

[0195] As one implementation method of this embodiment, scheduling instructions are executed, and the actual execution effect of the power generation equipment is continuously monitored in future time steps. The clearing capacity scoring sequence is continuously updated based on feedback data during execution.

[0196] As an optional approach in this embodiment, based on the equipment operation adjustment plan, the generated scheduling instructions can also be optimized using optimization algorithms (such as genetic algorithms, particle swarm optimization, etc.). By comparing multiple scheduling instruction schemes, the optimal scheme is selected for implementation to ensure maximum clearing capacity and guarantee the safe and stable operation of the power grid. During the optimization process, constraints such as the maximum output capacity of the equipment, the load constraints of the power grid, and the health status of the generating equipment need to be considered to ensure that the generated scheduling instructions are feasible.

[0197] This invention not only makes predictions but also possesses closed-loop feedback and proactive control capabilities. By conducting source analysis on low-scoring (risk) time steps, it pinpoints specific root cause indicators (such as a decline in power generation capacity), achieving a leap from "risk warning" to "cause diagnosis." Furthermore, based on the diagnostic results and combined with real-time status, it generates specific equipment operation adjustment plans and dispatch instructions, forming a complete closed loop of "assessment-prediction-early warning-diagnosis-control." This enables the system to proactively respond to risks, optimize power generation plans, thereby improving the grid friendliness and market competitiveness of new energy sources and ensuring the safe and stable operation of the power system.

[0198] Example 2

[0199] As another embodiment of this application, a power generation enterprise clearing capacity assessment system based on multi-indicator fusion is provided, employing the specific implementation methods disclosed in the above embodiments. The system includes a multi-indicator calculation module, a single-step graph modeling module, a clearing capacity assessment module, and a feedback adjustment module.

[0200] Multi-indicator calculation module: acquires equipment operation data, dispatch response data, and environmental operation parameters of new energy power generation enterprises, calculates multi-dimensional evaluation indicators that characterize the clearing ability of new energy power generation enterprises, and inputs the multi-dimensional evaluation indicators into the single-step graph modeling module;

[0201] Single-step graph modeling module: Constructs a graph structure using multi-dimensional evaluation indicators as nodes and determines the edge directions. Updates the edge weights for each time step and generates a single-step comprehensive graph representation for each time step. Inputs the single-step comprehensive graph representation into the clearing capability evaluation module.

[0202] Clearing Capacity Assessment Module: The single-step integrated graph representation is converted into a multi-step integrated graph representation sequence according to the time sequence, and input into the graph neural network prediction module integrated inside it to obtain the probability value of meeting the grid dispatch acceptance conditions in the future time step. The probability value is used as the clearing capacity score, and the clearing capacity score is input into the feedback adjustment module.

[0203] Feedback and Adjustment Module: Identifies risk time steps based on clearing capability scores, analyzes the causes of risks by combining the index values ​​of previous time steps, generates equipment operation adjustment plans, and outputs scheduling instructions for future time steps.

[0204] Example 3

[0205] This invention also proposes a terminal for implementing the power generation enterprise clearing capacity assessment method based on multi-indicator fusion described in Embodiment 1 above. The terminal includes a processor and a storage medium.

[0206] The storage medium is used to store instructions and data, including: equipment operation data, dispatch response data, and environmental operation parameters of new energy power generation enterprises; multi-dimensional evaluation index values ​​for each time step generated by the multi-index calculation module; single-step comprehensive graph representations for each time step generated by the single-step graph modeling module; model parameters and network structure definitions of the trained graph neural network prediction module; and preset operation parameters such as risk thresholds, sliding window size, and smoothing factors.

[0207] The processor is configured to operate according to the instructions to perform the steps according to the method described above.

[0208] The terminal can be deployed at the site or control center of a new energy power generation enterprise in the form of industrial-grade embedded devices, industrial control computers or servers, so as to realize localized real-time assessment and decision support.

[0209] Example 4

[0210] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the power generation enterprise clearing capacity assessment method based on multi-indicator fusion described in Embodiment 1 above.

[0211] Example 5

[0212] To verify the effectiveness of this invention, a simulation environment was constructed based on the operational data, scheduling records, and NWP meteorological data of a 200MW photovoltaic power plant over the past year (15-minute time step, 35,040 samples). The first 80% was used as the training set, and the last 20% as the test set. The model parameters were set as follows: 2 GCN layers, 64 hidden dimensions; bidirectional GRU (128 hidden dimensions) for the time-series coding layer; sliding window K=24 (covering the first 6 hours); smoothing factor α=0.8; risk threshold THᵣ=0.7.

[0213] Three typical comparison methods were selected:

[0214] Comparison 1 (Static Weighting Method): The static score is obtained by directly weighting and summing the five-dimensional indicators, without graphs or time series modeling;

[0215] Comparison 2 (LSTM time series method): The time series of the five-dimensional indicators are concatenated and input into LSTM for prediction, without building a graph structure;

[0216] Comparison 3 (Static Graph GNN method): Construct the same five-node graph but with fixed edge weights (no dynamic correlation adjustment).

[0217] The evaluation metrics used were Normalized Mean Absolute Error (NMAE), Normalized Root Mean Square Error (NRMSE), and Risk Time Step Warning Recall (the proportion of time steps that correctly identify actual risks). The experimental results are as follows:

[0218] Table 2 Comparison of experimental results

[0219]

[0220] Experimental results show that the present invention achieves lower normalized prediction errors (NMAE and NRMSE) than the comparative methods, and its risk warning recall rate reaches 91.7%, which is superior to static and ordinary time series models. This verifies that dynamic edge weight adjustment and time series graph modeling can effectively capture the coupling evolution law between multi-dimensional indicators, improve the prediction accuracy and risk identification ability of clearing capacity scoring, and provide a reliable basis for forward-looking scheduling decisions.

[0221] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0222] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0223] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0224] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0225] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for assessing the clearing-out capacity of power generation enterprises based on multi-indicator fusion, characterized in that, include: Step S1: Obtain equipment operation data, dispatch response data, and environmental operation parameters of new energy power generation enterprises, and calculate multi-dimensional evaluation indicators that characterize the clearing-out ability of new energy power generation enterprises; Step S2: Using multidimensional evaluation metrics as nodes, construct a graph structure and determine edge directions, update the edge weights at each time step, and generate a single-step comprehensive graph representation for each time step; Step S3: The single-step integrated graph representation is converted into a multi-step integrated graph representation sequence according to the time sequence, and the sequence is input into the graph neural network prediction module to obtain the probability value of meeting the grid dispatch acceptance conditions in the future time step, and the probability value is used as the clearing capability score. Step S4: Identify the risk time step based on the clearing capability score, analyze the cause of the risk by combining the index values ​​of the previous time step, generate the equipment operation adjustment plan, and output the scheduling instructions for the future time step.

2. The method for assessing the clearing-out capacity of power generation enterprises based on multi-indicator fusion according to claim 1, characterized in that, The multi-dimensional evaluation indicators characterizing the clearing-out ability of new energy power generation enterprises in step S1 include: The power generation capacity is calculated based on the rated power model of the power generation equipment, the equipment health factor, and the environmental correction coefficient. The equipment health factor is calculated by weighting the reliability value, the performance surplus coefficient, and the set positive adjustment term. The power generation stability index is calculated based on the real-time power output sequence of the power generation equipment, the fault impact factor, and the capacity difference. The execution scheduling capability is calculated based on the scheduling response delay, stabilization time, output error, and instruction type after the scheduling instruction is issued. The power generation forecast accuracy is calculated based on the mean absolute error and root mean square error between the power generation forecast data and the actual data, after normalization and weighting. The performance adaptability index is calculated based on the performance loss time and recovery time under disturbance conditions, after normalization and weighting.

3. The method for assessing the clearing capacity of power generation enterprises based on multi-indicator fusion according to claim 1, characterized in that: In step S2, the multidimensional evaluation index is used as nodes to construct a graph structure and the edge directions are determined. The edge weights at each time step are updated, specifically as follows: Multidimensional evaluation indicators are used as nodes to construct a graph structure, and the edge directions are determined based on the physical, scheduling, and environmental impact relationships among the multidimensional evaluation indicators. For two nodes with an edge, the node correlation is calculated using a sliding window method based on the historical index values ​​of the nodes. The edge weights at the current time step are dynamically adjusted by weighting the edge weights of the previous time step with the correlation using a smoothing factor.

4. The method for assessing the clearing-out capacity of power generation enterprises based on multi-indicator fusion according to claim 3, characterized in that: The step S2, which generates a single-step integrated diagram representation for each time step, specifically involves: For each time step, the information of the current node and its neighboring nodes is aggregated, and combined with the node features of historical time steps, to generate the representation vector of the current node at the current time step; At each time step, a single-step comprehensive graph representation is generated by aggregating the results from the graph convolutional network and dynamically adjusting the edge weights. The single-step synthetic graph representation includes node representation vectors and the updated edge weight matrix.

5. The method for assessing the clearing-out capacity of power generation enterprises based on multi-indicator fusion according to claim 1, characterized in that: In step S3, the multi-step integrated graph representation sequence, constructed from the single-step integrated graph representation, is input into the graph neural network prediction module, specifically as follows: The single-step synthesis graph representations output from multiple time steps are arranged in chronological order to form a multi-step synthesis graph representation sequence; The multi-step integrated graph representation sequence is input into the temporal coding layer to extract evolutionary features across time steps and output the hidden state of each node in the time step sequence. Global pooling is performed on the hidden states of all nodes to obtain a global graph representation.

6. The method for assessing the clearing capacity of power generation enterprises based on multi-indicator fusion according to claim 5, characterized in that: In step S3, the probability value of satisfying the grid dispatch acceptance condition in the future time step is obtained, and the probability value is used as the clearing capability score, specifically: The global graph representation is input into the fully connected layer, and the output is the probability value of the new energy power generation enterprise meeting the grid dispatch acceptance conditions in the future time step. The probability value is used as the clearing capability score.

7. The method for assessing the clearing capacity of power generation enterprises based on multi-indicator fusion according to claim 1, characterized in that: In step S4, the risk time step is identified based on the clearing capacity score, and the causes of risk are analyzed in conjunction with the indicator values ​​of the preceding time step. Specifically: The continuously output clearing capacity score is compared with the set threshold. If the clearing capability score output at any given time is lower than the set threshold, then... The time step is marked as a risk time step; Retrieve Time before The indicator values ​​at each time step are analyzed, and the trend of change of the indicator values ​​is analyzed using time series analysis to determine the reasons for the decline in clearing capacity.

8. The method for assessing the clearing capacity of power generation enterprises based on multi-indicator fusion according to claim 7, characterized in that: In step S4, generating a device operation adjustment plan and outputting scheduling instructions for future time steps specifically involves: Based on the identified risk causes, and using a pre-established cause-and-response mapping rule base, candidate adjustment strategies are matched for different risk causes. Taking into account the current status of power generation equipment, meteorological conditions, and grid operation requirements, the final adjustment strategy is selected from the candidate adjustment strategies. Based on the final adjustment strategy, a device operation adjustment plan is generated, and based on the preset mapping relationship, the device operation adjustment plan is converted into scheduling instructions.

9. A power generation enterprise clearing capacity assessment system based on multi-indicator fusion, characterized in that, It includes a multi-indicator calculation module, a single-step graph modeling module, a clearing capacity assessment module, and a feedback adjustment module: Multi-indicator calculation module: acquires equipment operation data, dispatch response data, and environmental operation parameters of new energy power generation enterprises, and calculates multi-dimensional evaluation indicators that characterize the clearing ability of new energy power generation enterprises; Single-step graph modeling module: Constructs a graph structure using multi-dimensional evaluation metrics as nodes and determines edge directions, updates the edge weights at each time step, and generates a single-step comprehensive graph representation for each time step; Clearing Capability Assessment Module: The single-step integrated graph representation is converted into a multi-step integrated graph representation sequence according to the time sequence. The sequence is then input into the graph neural network prediction module integrated within the module to obtain the probability value that the grid dispatch acceptance conditions will be met in the future time step. The probability value is used as the clearing capability score. Feedback and Adjustment Module: Identifies risk time steps based on clearing capability scores, analyzes the causes of risks by combining the index values ​​of previous time steps, generates equipment operation adjustment plans, and outputs scheduling instructions for future time steps.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it performs the steps of the method according to any one of claims 1-8.

Images