A cross-regional virtual power plant frequency regulation resource coordinated scheduling optimization method based on large model

Abstract

The present application relates to the technical field of power plant resource scheduling, and provides a cross-regional virtual power plant frequency regulation resource collaborative scheduling optimization method combined with a large model. The method includes: establishing a cross-regional resource pool and labeling it, establishing a resource sharing architecture; calling high-frequency sensors to monitor the load and frequency of each region, and establishing a real-time demand data set; activating the adversarial network to expand the historical data set of each region, performing time series prediction of the real-time demand data set, and establishing a time series prediction result; generating a connection stability value based on the connected identifier, and performing collaborative scheduling optimization based on the time series prediction result, label identification, and connection stability value to generate an optimization result; reconstructing the resource sharing architecture based on the optimization result, and performing collaborative scheduling management of power plant frequency regulation resources to solve the technical problems of low resource utilization and difficulty in collaborative scheduling between regions due to uneven resource distribution and inconsistent response speed in each region, so as to achieve the effect of improving the collaborative response speed and resource utilization of frequency regulation resources.

Description

Technical Field

[0001] The present application relates to the technical field of power system management, specifically to the technical field of power plant resource scheduling, and in particular to a method for optimizing the coordinated scheduling of frequency regulation resources of cross-regional virtual power plants combined with a large model. Background Art

[0002] In the current development of the power system, efficient dispatching of power resources and frequency stability management have become key issues. With the continuous expansion of the scale of the power grid and the complexity of power load demand, the traditional power plant frequency dispatching method has gradually exposed the defects of insufficient resource integration, delayed dispatching response, poor frequency control accuracy, etc. when dealing with dispatching tasks with wide regional distribution, high resource heterogeneity and strong real-time response requirements. It is difficult to meet the needs of modern power systems for resource utilization efficiency and grid frequency stability. The existing technology mainly relies on the frequency regulation resources of distributed power plants in the region for dispatching, but when faced with cross-regional frequency regulation needs, it is difficult to achieve dynamic integration and efficient utilization of resources, resulting in the inability to achieve rapid coordinated dispatching of frequency regulation resources between different regions. Summary of the invention

[0003] This application provides a cross-regional virtual power plant frequency regulation resource coordinated scheduling optimization method combined with a large model, aiming to solve the technical problems of low resource utilization and difficult coordinated scheduling between regions due to uneven resource distribution and inconsistent response speed in each region. It achieves the technical effect of improving the coordinated response speed and resource utilization of frequency regulation resources by building a cross-regional virtual resource pool and a multi-dimensional resource sharing architecture, thereby ensuring the stability of the power grid frequency and the flexibility of frequency regulation management.

[0004] In view of the above problems, the present application provides a method for coordinated scheduling and optimization of frequency regulation resources of cross-regional virtual power plants combined with a large model.

[0005] The present application provides a method for collaborative scheduling optimization of frequency regulation resources of a cross-regional virtual power plant combined with a large model, the method comprising: establishing a cross-regional resource pool, the cross-regional resource pool being constructed by integrating and virtualizing the frequency regulation resources of each region, and labeling the cross-regional resource pool to establish a multi-dimensional resource sharing architecture, the label identification comprising a connected identification and a resource label identification; calling high-frequency sensors to monitor the load and frequency of each region and establish a real-time demand data set; activating an adversarial network to perform data expansion on the historical data set of each region, performing time series demand prediction of the real-time demand data set, and establishing a time series prediction result, the time series prediction result comprising a short-term prediction result and a long-term prediction result; generating a connection stability value of the multi-dimensional resource sharing architecture according to the connected identification, performing collaborative scheduling optimization based on the time series prediction result, the label identification, and the connection stability value, and establishing a collaborative scheduling optimization result; reconstructing the multi-dimensional resource sharing architecture according to the collaborative scheduling optimization result, and using the reconstructed multi-dimensional resource sharing architecture to perform collaborative scheduling management of power plant frequency regulation resources.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] The above cross-regional virtual power plant frequency regulation resource collaborative dispatch optimization method combined with a large model first integrates the frequency regulation resources of each region to establish a cross-regional resource pool, virtualizes the frequency regulation resources distributed in different regions, and classifies and manages them through label identification to build a resource architecture with multi-dimensional sharing capabilities. The construction of this resource pool improves the efficiency and flexibility of resource integration, so that resources between regions can be more quickly coordinated. Subsequently, the load and frequency changes of each region are monitored in real time by high-frequency sensors to generate an instant demand data set, thereby ensuring the real-time and accuracy of the frequency regulation response. After that, the adversarial network is activated to expand the historical data set of each region to make the data richer and more complete, and enhance the accuracy and comprehensiveness of the time series demand forecast. On this basis, the time series forecast of the demand data is performed to generate short-term and long-term forecast results, so as to more accurately grasp the changing trend of the frequency regulation demand. At the same time, the connection stability value of the resource architecture is calculated by using the connected identifier, and the optimization of the collaborative dispatch is carried out in combination with the time series forecast results and label identification to establish a better dispatching scheme. The collaborative dispatch optimization process can effectively balance the resource demand and response of different regions and avoid frequency fluctuations and resource waste. Finally, the resource sharing architecture is reconstructed based on the optimization results, and the optimized architecture is used to coordinate the dispatch and management of the frequency regulation resources of the power plant, further improving the overall response speed and resource utilization efficiency of the frequency regulation process, and ensuring the stability and safety of the power grid frequency.

[0008] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 It is a flow chart of a method for collaborative scheduling and optimization of frequency regulation resources of cross-regional virtual power plants combined with a large model in an embodiment.

[0011] Figure 2 It is a flowchart of establishing collaborative scheduling optimization results of a method for collaborative scheduling optimization of frequency regulation resources of cross-regional virtual power plants combined with a large model in an embodiment. DETAILED DESCRIPTION

[0012] The embodiments of the present application provide a cross-regional virtual power plant frequency regulation resource coordinated scheduling optimization method combined with a large model to solve the technical problems of low resource utilization and difficult coordinated scheduling between regions due to uneven resource distribution and inconsistent response speed in each region. It achieves the technical effect of improving the coordinated response speed and resource utilization of frequency regulation resources and ensuring the stability of power grid frequency and the flexibility of frequency regulation management by building a cross-regional virtual resource pool and a multi-dimensional resource sharing architecture.

[0013] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0014] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0015] Examples, such as Figure 1As shown, the present application provides a cross-regional virtual power plant frequency regulation resource coordinated scheduling optimization method combined with a large model, the method comprising:

[0016] A cross-regional resource pool is established. The cross-regional resource pool is constructed by integrating and virtualizing the frequency modulation resources of each region, and the cross-regional resource pool is labeled to establish a multi-dimensional resource sharing architecture. The label identification includes a connected identification and a resource label identification.

[0017] In an embodiment of the present application, the system terminal integrates and virtualizes the frequency modulation resources of each region, including unified resource abstraction and digital modeling of physical frequency modulation resources (such as power plants, energy storage facilities, etc.) in different regions, so that each resource can be managed in a unified virtual resource pool. Through virtualization, the physical resources of each region can be presented in a unified manner, which facilitates cross-regional coordinated calls. In order to facilitate the management and differentiation of resources, the system terminal assigns a label identifier to each virtualized frequency modulation resource. These label identifiers include at least the following types of information, namely, connected identifiers and resource label identifiers. Among them, the connected identifier is used to identify whether the current resource is in a connected state, ensuring that only resources in an available state are called to avoid scheduling errors caused by offline or maintenance. The resource label identifier is used to further classify resources, including information such as resource type, regional location, capacity, response time, etc. These identifiers provide a basis for subsequent resource screening and allocation. Based on the label identifier, a pre-built cross-regional resource sharing architecture interface is used to connect and integrate with the cross-regional resource pool that has completed the label identifier to establish a multi-dimensional resource sharing architecture. This multi-dimensional resource sharing architecture allows resources to be divided and managed according to frequency modulation requirements in different dimensions, such as resource type, regional location, capacity, response speed, etc. Through this architecture, system terminals can flexibly call resources that meet different frequency modulation requirements based on current load and frequency requirements, and realize efficient sharing and coordination of cross-regional resources.

[0018] High-frequency sensors are used to monitor the load and frequency in each area and establish a real-time demand data set.

[0019] In one embodiment, the system terminal monitors the power load and frequency fluctuation data of each area in real time through high-frequency sensors deployed in various areas. High-frequency sensors can collect data frequently in a short period of time to ensure that even the slightest fluctuations in load and frequency changes can be captured. By continuously collecting and recording these data, a data set containing real-time demand information is gradually constructed to provide accurate and real-time load demand basis for subsequent frequency regulation decisions. The establishment of this real-time demand data set enables the system terminal to respond more quickly to load changes in the area and ensure the stability and accuracy of the frequency regulation process.

[0020] After activating the adversarial network to perform data expansion on the historical data sets of each region, the time series demand forecasting of the real-time demand data set is performed to establish a time series forecasting result, which includes a short-term forecasting result and a long-term forecasting result.

[0021] In one embodiment, in order to make the prediction of real-time demand more accurate, the system terminal activates the alternating optimization process of the adversarial network. In this optimization process, the missing data will be defined based on the historical data set, and the generative network and the discriminative network in the adversarial network will be alternately trained based on the missing data and the historical data set, so that the adversarial network can be exposed to more comprehensive and diverse data, improve the generalization ability of the adversarial network, and understand the patterns of load and frequency changes more deeply and comprehensively. After completing the data expansion, the system terminal uses this expanded historical data set to perform time series demand forecasting on the real-time demand data set, and generates time series forecasting results including short-term and long-term. The short-term forecast results can help the system terminal quickly respond to the upcoming load fluctuations and ensure the efficient response of frequency modulation resources; while the long-term forecast results provide support for future resource allocation and scheduling strategies, and improve the foresight and stability of scheduling.

[0022] Furthermore, the present application provides that the activation adversarial network performs data expansion on the historical data sets of each region, and further includes:

[0023] After data cleaning of the historical data set, data richness missing analysis is performed to establish the data demand space, and the data boundary is defined by the historical data set; the guided noise is established based on the data demand space, and the cleaned historical data set is used as the basic data to generate data for the generative network based on the basic data and guided noise; the discriminant network is initialized through the data boundary, and the generated data is discriminated by the initialized discriminant network; the alternating optimization of the generative network and the discriminant network is performed to complete data expansion.

[0024] Preferably, the system terminal performs data cleaning on the historical data set, identifies and removes outliers beyond the normal range through statistical methods (such as Z-score or IQR method), and ensures the independence of data samples by checking and removing duplicates in the historical data set. After cleaning, data richness missing analysis is performed, that is, the distribution and density of existing data in the cleaned historical data set in dimensions such as time, geographical location, and load type are statistically analyzed to identify blank and insufficient areas in the data, and thereby construct a data demand space, that is, the target range required for data generation. For example, if the load data on weekdays is relatively complete through the distribution of statistical data in time, but the load data on weekends is less or even missing, then these weekend periods constitute a blank area in the time dimension. Subsequently, by analyzing the cleaned historical data set, boundary values ​​are set according to data such as load peak and minimum load, and data boundaries are determined to prevent non-actual abnormal data from appearing in the subsequent data generation process. This data boundary can ensure that the generated data is within the actual operating range and avoid the generation of meaningless or invalid data samples. Afterwards, based on the data demand space, guide noise for generating network design. Guided noise is used to guide the generative network to generate data samples that meet the target characteristics during the data generation process. The cleaned historical data is used as the basic data and combined with the guided noise to provide rich and diverse inputs for the generative network so as to generate load or frequency data under different conditions. Before generating data, the discriminant network is initialized to ensure that it can identify the authenticity and validity of the generated data. By incorporating data boundary information into the initialization process, the discriminant network has basic judgment capabilities and can identify whether the generated data is within a reasonable range. This initialization process provides a basis for the alternating optimization of the generative network and the discriminant network. Then, in the data generation phase, the generative network uses the basic data and guided noise to generate new data samples, which are then passed to the discriminant network for verification. The discriminant network determines the similarity between the generated data and the real data by calculating the cross entropy loss function, thereby determining whether the generated data meets the characteristics of the actual historical data. If the discrimination result shows that the generated data does not meet the real characteristics, the discriminant network will feedback information to the generative network to prompt it to improve the quality of the generated data. According to the feedback from the discriminant network, the generative network adjusts its parameters by calculating the adversarial loss function of the generative network, so that the generated data gradually approaches the real data distribution. In this process, the generative network and the discriminative network are optimized in an alternating manner, that is, the parameters of the discriminative network are optimized first to make it more accurately distinguish between real data and generated data, and then the parameters of the discriminative network are fixed and the parameters of the generative network are optimized to make the generated data closer to the real data distribution. Through multiple rounds of such alternating training, the generative network gradually improves the authenticity and diversity of the generated data, making the generated data samples gradually approach the real historical data characteristics.The fully trained generation network can generate high-quality load or frequency data that conforms to the characteristics of historical data, completing the expansion of historical data sets. These generated data can effectively fill in the data gaps and insufficient areas in dimensions such as time, geographic location, and load type, providing more comprehensive support for time series demand forecasting.

[0025] Furthermore, the present application provides the time series demand prediction of the execution real-time demand data set, and establishes the time series prediction result, and also includes:

[0026] Configure the window space of short time series windows and long time series windows; use the real-time demand data set to reconstruct the short time series windows and long time series windows in the window space, and establish the reconstruction results; use the reconstruction results to predict the demand similarity of the real-time demand data set based on the historical data set after data expansion, and establish the time series prediction results.

[0027] Preferably, in order to predict the real-time demand, the system terminal configures a window space of a short time series window and a long time series window. The short time series window and the long time series window represent the observation range of different time spans of the data respectively. The short time series window is used to capture short-term fluctuations and detailed changes, while the long time series window is used to identify long-term trends and patterns. The system terminal sets the lengths of the two windows according to the forecast demand. For example, the short time series window may be from a few minutes to a few hours, while the long time series window is from a few days to a few weeks. Through reasonable window configuration, it can be ensured that the change characteristics of the data can be captured at different time scales. Subsequently, the short time series window and the long time series window are reconstructed in the window space using the real-time demand data set. Data points are extracted in each window, and subsets of different time scales are reconstructed according to the time interval of the data, so that the subsequent prediction process can better analyze the demand changes. For example, in the short time series window, the real-time demand data set will be decomposed into load change data of the corresponding time interval according to the window space of the short time series window, and in the long time series window, the same operation will be performed to form short-term and long-term demand data structures, that is, reconstruction results. The reconstruction results integrate the data features at different time scales, so that the subsequent time series demand forecasting model can more comprehensively capture the short-term and long-term dynamic changes of real-time demand. Afterwards, the reconstruction results are used to predict the similarity of the demand of the real-time demand data set based on the historical data set after data expansion. In this step, the system terminal performs the same operation on the expanded historical data set according to the window space of the short time series window and the long time series window, determines the historical reconstruction results of the historical data set, and then trains the time series demand forecasting model based on this historical reconstruction result. After the training is completed, the reconstruction results are input into the time series demand forecasting model for matching analysis, finding similar demand patterns, predicting future demand changes, and generating time series forecasting results including short-term and long-term. These forecast results can reflect possible future load demand fluctuations and provide guidance for resource scheduling.

[0028] For the training of the time series demand forecasting model, the system terminal divides the historical load data of short and long time series in the historical reconstruction results into a training set and a validation set, and trains the constructed time series forecasting model through the training set. After the training, the validation set is used for verification to evaluate the prediction accuracy of the model. Among them, the time series forecasting model can be constructed based on models such as LSTM, RNN or Transformer. Taking the LSTM model as an example, a time series forecasting model structure is constructed, including an input layer, multiple LSTM layers, a fully connected layer and an output layer. Then, the weights of the time series forecasting model are initialized using random numbers, and the training set is input into the initialized forecasting model for forward propagation. The short-term and long-term load forecasting results are calculated by passing through the input layer, LSTM layer, fully connected layer and output layer layer by layer. Subsequently, the mean square error loss function is used to calculate the loss value between the prediction result and the real data, and the gradient of the loss to the weight of each layer is calculated layer by layer through back propagation. Then, the Adam optimizer is used to optimize the model parameters and adjust the weights to minimize the value of the loss function. The above process is repeated until the maximum number of iterations is reached or the loss function converges. After the training is completed, the model performance is tested using the validation set to evaluate the accuracy of the model in short-term and long-term load forecasting. If the forecast accuracy meets expectations, the current time series forecasting model is output as the final model. Otherwise, hyperparameters such as the learning rate and the number of training batches are adjusted to further improve the model's forecasting effect.

[0029] The connection stability value of the multi-dimensional resource sharing architecture is generated according to the connected identifier, and the collaborative scheduling optimization is performed based on the timing prediction results, label identifiers, and connection stability value to establish the collaborative scheduling optimization result.

[0030] In one embodiment, the system terminal adds 1 to the interruption frequency of the resource with the connected identifier and performs a reciprocal calculation to obtain the connection stability value of each resource, and then performs an average calculation to obtain the connection stability value in the current multi-dimensional resource sharing architecture. The connection stability value is used to measure the connection status and availability stability of the resource, ensuring that those resources with stable status and reliable response are given priority during collaborative scheduling. Subsequently, the system terminal inputs the timing prediction results, label identifiers, and connection stability values ​​into the collaborative scheduling optimization network for collaborative scheduling optimization. The timing prediction results provide short-term and long-term load demand trends, the label identifiers help screen out resources that meet specific needs, and the connection stability value ensures the reliability of the scheduling scheme. During the optimization process, the optimal resource allocation and call scheme is determined by comprehensively analyzing the processing results of each level in the collaborative scheduling optimization network to achieve overall balance and efficiency maximization of the scheduling. Finally, the generated collaborative scheduling optimization results reflect the best scheduling scheme under the current resource state, providing support for resource sharing and load balancing.

[0031] Further, if Figure 2 As shown, the present application provides the collaborative scheduling optimization based on the timing prediction result, the label identifier, and the connection stability value, and establishes the collaborative scheduling optimization result, and also includes:

[0032] A collaborative scheduling optimization network is configured, and after receiving the label identification at the basic processing layer of the collaborative scheduling optimization network, the environmental protection coefficient of each resource is parsed and obtained; the collaborative evaluation function of the collaborative scheduling optimization network is activated, and a collaborative scheduling evaluation is performed based on the environmental protection coefficient and the timing prediction result, and a collaborative scheduling evaluation result is established, wherein the collaborative scheduling evaluation includes an environmental protection evaluation, a response effect evaluation, a cost evaluation, and a long-short timing adaptation evaluation; a penalty processing layer is used to perform a destruction penalty analysis on the collaborative scheduling scheme corresponding to the collaborative scheduling evaluation result based on the connection stability value, and establish a destruction penalty coefficient; the collaborative scheduling evaluation result is penalized and corrected with the destruction penalty coefficient, and a collaborative scheduling optimization result is established.

[0033] Preferably, the system terminal is configured with a collaborative scheduling optimization network, which includes an input layer, a basic processing layer, a collaborative evaluation layer, a penalty processing layer, an output layer, etc. The input layer is used to receive the input timing prediction results, label identification, and connection stability values, and pass them to subsequent layers. The basic processing layer is used to parse out the environmental protection coefficient of each resource. The collaborative evaluation layer includes a collaborative evaluation function for parsing the environmental protection coefficient and timing prediction results. The penalty processing layer is used to correct the collaborative scheduling evaluation results. The output layer is used to output the corrected collaborative scheduling evaluation results. In the basic processing layer of the network, the label identification information of each resource transmitted by the input layer is received, and the environmental protection coefficient of each resource is calculated based on the analysis of the environmental protection related information in the label identification information, such as carbon emissions, energy efficiency, etc. This basic processing layer has been pre-trained with samples, and the training method is the same as above. The environmental protection coefficient is used to measure the degree of environmental impact of resources and provide basic data on environmental protection for subsequent collaborative scheduling evaluation. Subsequently, the calculated environmental protection coefficient and the input timing prediction results are passed to the collaborative evaluation layer, and the internal collaborative evaluation function is activated to analyze the environmental protection coefficient and timing prediction results. The environmental protection evaluation is obtained by calculating the mean value of the environmental protection coefficient of each resource in the collaborative scheduling scheme; the response effect evaluation is obtained by evaluating the response time and response accuracy of the collaborative scheduling scheme; the cost evaluation is obtained by cost analysis of the collaborative scheduling scheme; and the long-short timing adaptation evaluation is obtained by consistency evaluation of the collaborative scheduling scheme. By summarizing these evaluations, a collaborative scheduling evaluation is formed. After that, the penalty processing layer is entered to perform a damage penalty analysis based on the connection stability value for the current collaborative scheduling scheme. By comparing the connection stability value with whether there are resources with unstable connection status in the current scheme. If the connection stability value of some resources is low, it will have a negative impact on the scheduling effect. The penalty processing layer will generate corresponding damage penalty coefficients according to the proportion of these unstable resources as a correction factor for the evaluation results. Finally, the damage penalty coefficient is used to penalize the collaborative scheduling evaluation results, and the penalty coefficient is applied to the preliminary scheduling evaluation results to weaken the impact of unstable resources on the scheduling scheme and ensure the reliability and applicability of the scheme. After correction, the system terminal generates the final collaborative scheduling optimization result and outputs an optimal scheduling plan that balances environmental protection, responsiveness, cost and adaptability.

[0034] Furthermore, the present application provides that the response effect evaluation also includes:

[0035] Based on the collaborative scheduling scheme, the response time is fitted for the timing prediction results, and a response speed feedback evaluation is established according to the response time fitting results. The frequency recovery rate and frequency deviation control of the collaborative scheduling scheme are calculated, and a response accuracy feedback evaluation is established according to the calculation results. The stability of the collaborative scheduling scheme is analyzed, and a response stability feedback evaluation is established. The response effect evaluation is generated by the response speed feedback evaluation, the response accuracy feedback evaluation, and the response stability feedback evaluation.

[0036] Optionally, the system terminal performs response time fitting on the timing prediction results based on the coordinated scheduling scheme. According to the predicted load demand and the simulated execution of the coordinated scheduling scheme, the response speed of the resource under load fluctuation is obtained. For example, the time delay from the change of load demand to the start of the response of the scheduling scheme is recorded, and the response time fitting value is calculated. According to the response time fitting result, the response speed of the scheduling scheme is fed back and evaluated. The faster the response speed, the higher the evaluation, which can ensure the real-time performance of the frequency modulation scheme. In addition, the frequency recovery rate and frequency deviation control calculation are performed on the coordinated scheduling scheme to establish a response accuracy feedback evaluation. The frequency recovery rate refers to the recovery effect of the scheduling scheme on the system frequency after the load demand fluctuates, which is measured by the speed at which the frequency recovers to the target range. The frequency deviation control is to calculate the deviation of the system frequency from the target frequency during the execution of the scheduling scheme. The response accuracy feedback evaluation is established by calculating the results of the frequency recovery rate and frequency deviation control. The higher the recovery rate and the smaller the deviation, the more accurately the scheduling scheme can match the load demand and the higher the response accuracy evaluation. In addition, the scheme stability analysis is performed on the coordinated scheduling scheme to establish a response stability feedback evaluation. The performance of the system terminal in tracking the scheduling scheme in multiple time periods is measured by weighting the number of disconnections and the number of response fluctuations greater than the preset fluctuation constraint, and then adding it to 1 for inverse calculation. If the scheduling scheme shows a stable and continuous frequency regulation effect under load changes, that is, there are no frequent disconnections or large response fluctuations, the stability evaluation will be higher. Finally, the various feedback evaluations are integrated to generate a response effect evaluation. Through weighted calculation (the weights are determined according to actual needs) combined with the response speed feedback evaluation, response accuracy feedback evaluation and response stability feedback evaluation, a comprehensive response effect evaluation of the scheduling scheme is formed, reflecting the overall frequency regulation effect of the scheme under actual load fluctuation scenarios, and providing a basis for the final scheduling decision.

[0037] Furthermore, the present application provides that the cost evaluation also includes:

[0038] Conduct cost analysis of frequency modulation resources of energy types under the collaborative scheduling scheme based on label identification to establish an energy cost evaluation. Conduct distance-based cost analysis based on cross-regional connections of the collaborative scheduling scheme to establish a scheduling cost evaluation. Conduct attenuation cost analysis of frequency modulation resources for the collaborative scheduling scheme to establish an attenuation cost evaluation. Generate a cost evaluation based on energy cost evaluation, scheduling cost evaluation, and attenuation cost evaluation.

[0039] Optionally, the system terminal extracts the energy type (such as thermal power, wind power, photovoltaic power, energy storage, etc.) of each resource and its corresponding unit cost from the resource tag identifier, and calculates the overall energy cost based on the different energy types and their usage ratios in the scheduling scheme. By summarizing the unit cost and scheduling amount of each energy, an energy cost evaluation is formed to measure the energy usage economy of the scheme. During the scheduling process, cross-regional calls of resources in different regions may incur additional transmission and scheduling costs. The system terminal determines the transmission distance in the scheduling process by calculating the distance between the geographical location of the resource and the load center. The scheduling cost is usually proportional to the transmission distance, and the longer the distance, the higher the cost. Through the proportional relationship between the distance and the transmission cost, the system terminal calculates the scheduling cost of the cross-regional connection and establishes a scheduling cost evaluation. The use of some resources (such as energy storage equipment) is accompanied by certain performance degradation or loss. For example, energy storage equipment will experience capacity degradation during frequent charging and discharging, and traditional power generation equipment may reduce efficiency due to high load operation. Based on the decay parameters in the resource tags, the system terminal evaluates the decay of each resource in the scheduling scheme and calculates the additional cost caused by decay, that is, multiplying the decay parameters by the unit maintenance cost or replacement cost. The selection of unit maintenance cost and replacement cost depends on whether the decay parameters reach the replacement threshold. The higher the decay cost evaluation, the more frequent resource replenishment or maintenance the scheme may require, and the higher the cost. Finally, the overall cost evaluation is generated by combining the energy cost evaluation, the scheduling cost evaluation, and the decay cost evaluation in the same way. This cost evaluation is used to comprehensively measure the economic feasibility of the collaborative scheduling scheme and provide support for scheduling decisions. In scheduling optimization, schemes with lower cost evaluation scores have more economic advantages.

[0040] Furthermore, the present application provides that the long-short timing adaptation evaluation also includes:

[0041] Analyze the collaborative scheduling scheme, establish short-term collaborative scheduling scheme and long-term collaborative scheduling scheme; conduct scheme consistency evaluation on the short-term collaborative scheduling scheme and long-term collaborative scheduling scheme, and establish scheme consistency evaluation results; use the scheme consistency evaluation results to generate the long-term and short-term adaptation evaluation of the corresponding collaborative scheduling scheme.

[0042] Optionally, the system terminal divides the scheduling scheme into two parts: a short-term coordinated scheduling scheme and a long-term coordinated scheduling scheme according to the needs of the scheduling scheme. The short-term coordinated scheduling scheme focuses on meeting short-term demand fluctuations, such as frequency regulation and load response at the hourly level; while the long-term coordinated scheduling scheme focuses on coping with resource allocation requirements for a longer period of time, such as daily, weekly or monthly load balancing and resource planning. The system terminal determines the boundaries of the division according to the time range of the load demand, so that the short-term scheme mainly covers short-term demand fluctuations, and the long-term scheme reflects the resource strategy over a longer period of time. Subsequently, the short-term and long-term scheduling schemes are evaluated for consistency to establish the scheme consistency evaluation results. The goal of the scheme consistency evaluation is to ensure that the short-term and long-term schemes are consistent in resource selection and scheduling strategies to avoid resource conflicts and scheduling errors. The evaluation content includes resource usage consistency, scheduling strategy consistency, and load adaptation consistency. For resource usage consistency, check whether the short-term and long-term schemes have conflicting uses on the same resource, and count the number of repeated scheduling of the same resource. For scheduling strategy consistency, the system terminal evaluates whether the scheduling strategies of the short-term and long-term plans are consistent, and calculates the ratio of the total duration of calling different strategies in the same time period to the total duration of the long-term plan. For load adaptation consistency, evaluate whether the short-term and long-term plans match the same load changes, and count the number of mismatches. For example, if the short-term plan responds to load fluctuations, but the long-term plan ignores the long-term load trend, it is counted as one time. After that, the maximum-minimum method is used to normalize the number of repeated scheduling and the number of mismatches, and then the normalized number of repeated scheduling, the normalized number of mismatches, and the ratio of the total duration of calling different strategies to the total duration of the long-term plan are averaged to obtain the scheme consistency evaluation result. Finally, the system terminal uses the scheme consistency evaluation result as the long-short time adaptation evaluation. If the consistency evaluation is high, the long-short time sequence adaptation evaluation score is also high, indicating that the short-term and long-term time sequence schemes are well coordinated and suitable for dealing with complex load changes; if the consistency evaluation is low, the long-short time sequence adaptation evaluation score will also decrease, indicating that the scheme is insufficient in adapting to demands of different time scales and needs further optimization. Through this process, the long-short time sequence adaptation evaluation can fully reflect the coordination and matching degree of the coordinated scheduling scheme in short-term and long-term demand response, and provide a basis for the selection of the final scheduling scheme.

[0043] The multi-dimensional resource sharing architecture is reconstructed according to the collaborative dispatch optimization results, and the collaborative dispatch management of power plant frequency regulation resources is carried out based on the reconstructed multi-dimensional resource sharing architecture.

[0044] In one embodiment, the system terminal analyzes the collaborative scheduling optimization results and identifies the resource characteristics and allocation requirements required for scheduling. The collaborative scheduling optimization results provide an optimized resource configuration scheme, including key factors such as the resource type to be called first, the scheduling time window, the geographical location distribution, the response speed and the cost. The system terminal analyzes these scheduling requirements into specific configuration requirements in the multi-dimensional resource sharing architecture. Subsequently, the multi-dimensional resource sharing architecture is reconstructed to adapt to the optimized resource requirements. During the reconstruction process, the system terminal rearranges the priority of each resource based on the collaborative scheduling optimization results. For example, resources with low cost and fast response are called first, or clean energy resources are called first according to the environmental protection coefficient. The priority update ensures that the most suitable resource can be quickly retrieved during the scheduling process. According to the collaborative scheduling optimization results, the label identification is adjusted to better reflect the optimized resource requirements. For example, the labels of some resources are changed to high priority or short-term response for fast scheduling. According to the load adaptability of each resource in the collaborative scheduling optimization results, the load allocation upper limit or dynamic allocation strategy of the resource is set to ensure that the resource will not be overloaded or out of frequency regulation requirements during scheduling. Finally, the reconstructed multi-dimensional resource sharing architecture is used for the collaborative scheduling management of power plant frequency regulation resources. During the dispatching process, the system terminal calls the optimized resource allocation plan in real time, quickly retrieves and allocates resources through a multi-dimensional architecture, and realizes efficient collaborative frequency management. The reconstructed architecture can quickly respond to short-term load fluctuations while maintaining stable support for long-term dispatching, thereby ensuring the overall frequency regulation efficiency and stability of the power plant.

[0045] Furthermore, the method also includes

[0046] Record the collaborative dispatching data of power plant frequency regulation resources, establish a collaborative dispatching database, and identify the collaborative dispatching database based on dispatching feedback; use dispatching feedback to build deep collaborative associations, and update the multi-dimensional resource sharing architecture based on deep collaborative associations.

[0047] Preferably, in the collaborative dispatching process of the frequency regulation resources of the power plant, the system terminal will record the relevant data of each dispatch, including the time of resource call, response speed, load distribution, dispatching cost, frequency control effect, etc. All these data are summarized in the collaborative dispatching database to form a historical record and behavior archive of the dispatching data. The database provides basic data support for subsequent optimization analysis and performance improvement. Subsequently, the collaborative dispatching database is identified based on the dispatching feedback. The dispatching feedback includes key evaluation indicators such as the success rate, response delay, and frequency regulation effect in the dispatching process. The system terminal identifies the record of each dispatch according to the feedback results. For example, if the response speed of a certain dispatch is fast and the frequency regulation effect is good, it is identified as efficient; if the response delay is high or the frequency regulation effect is poor, it is identified as inefficient or needs to be optimized. These identifications help the system terminal identify resources and solutions that perform well or need to be improved in the dispatching, and provide a reference for future dispatching decisions. After that, the system terminal analyzes the historical dispatching data to explore the collaborative relationship between various resources. For example, if some resources always show efficient response effects under specific scheduling combinations, the combination of these resources will be recorded as positive deep collaborative associations; on the contrary, if some resource combinations perform poorly in historical records, negative collaborative associations will be formed. By analyzing these associations, we can better understand the collaborative effects of different resource combinations. Finally, the system terminal applies the information of deep collaborative associations to the update of the resource sharing architecture to improve the scheduling effect of the architecture. For example, prioritize resource combinations with positive collaborative associations for quick retrieval; or reduce the priority of negative collaborative association combinations to avoid inefficient scheduling. The system terminal dynamically adjusts the sharing architecture based on the latest collaborative association information, so that it gradually forms the optimal resource call structure, thereby achieving more efficient resource sharing and load response in future frequency regulation needs.

[0048] In summary, the embodiments of the present application have at least the following technical effects:

[0049] The embodiment of the present application integrates the frequency modulation resources of each region to build a virtualized cross-regional resource pool, assigns a label identifier to each resource, and then calls a high-frequency sensor to monitor the load and frequency changes in real time, establishes a real-time demand data set, and expands historical data through an adversarial network, so as to make short-term and long-term time series predictions for demand. On this basis, according to information such as label identifiers and connection stability values, the collaborative scheduling scheme of resources is optimized, and the multi-dimensional factors such as environmental protection, response effect, and cost of the scheme are evaluated through the collaborative scheduling optimization network, and then the scheme is penalized and corrected based on the connection stability. In addition, the method includes long-short time series adaptation evaluation, historical data cleaning and expansion, feedback identification of the scheduling database and deep collaborative association update, so as to continuously optimize the multi-dimensional resource sharing architecture, and finally realize the efficient collaborative management and scheduling decision of the frequency modulation resources of the power plant at different time scales. These technical effects jointly solve the technical problems of low resource utilization and difficulty in inter-regional collaborative scheduling due to uneven resource distribution and inconsistent response speed in each region, and realize the effect of improving the collaborative response speed and resource utilization of frequency modulation resources by building a cross-regional virtual resource pool and a multi-dimensional resource sharing architecture, ensuring the stability of the power grid frequency and the flexibility of frequency modulation management.

[0050] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0051] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0052] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A cross-regional virtual power plant frequency regulation resource coordinated scheduling optimization method combined with a large model, characterized in that: The method comprises: Establish a cross-regional resource pool, which is constructed by integrating and virtualizing the frequency modulation resources of each region, and labeling the cross-regional resource pool to establish a multi-dimensional resource sharing architecture, wherein the label identification includes a connected identification and a resource label identification; Use high-frequency sensors to monitor load and frequency in each area and establish a real-time demand data set; After activating the adversarial network to perform data expansion on the historical data sets of each region, the time series demand forecast of the real-time demand data set is performed to establish a time series forecast result, which includes a short-term forecast result and a long-term forecast result; Generate the connection stability value of the multi-dimensional resource sharing architecture based on the connected identifier, perform collaborative scheduling optimization based on the timing prediction results, label identifiers, and connection stability values, and establish the collaborative scheduling optimization results; The multi-dimensional resource sharing architecture is reconstructed according to the collaborative dispatch optimization results, and the collaborative dispatch management of power plant frequency regulation resources is carried out based on the reconstructed multi-dimensional resource sharing architecture.

2. The cross-regional virtual power plant frequency regulation resource coordinated scheduling optimization method combined with a large model as claimed in claim 1 is characterized in that: The collaborative scheduling optimization is performed based on the time series prediction result, the label identifier, and the connection stability value, and the collaborative scheduling optimization result is established, and further includes: Configure a collaborative scheduling optimization network, so that after receiving the label identification, the basic processing layer of the collaborative scheduling optimization network parses and obtains the environmental protection coefficient of each resource; Activate the collaborative evaluation function of the collaborative scheduling optimization network, perform collaborative scheduling evaluation based on the environmental protection coefficient and the time series prediction results, and establish a collaborative scheduling evaluation result, wherein the collaborative scheduling evaluation includes environmental protection evaluation, response effect evaluation, cost evaluation, and long-short time series adaptation evaluation; The penalty processing layer is used to perform a destruction penalty analysis on the collaborative scheduling scheme corresponding to the collaborative scheduling evaluation result based on the connection stability value and establish a destruction penalty coefficient; The collaborative scheduling evaluation results are penalized and corrected with the destruction penalty coefficient to establish the collaborative scheduling optimization result.

3. The cross-regional virtual power plant frequency regulation resource coordinated scheduling optimization method combined with a large model as claimed in claim 2 is characterized in that: The response effect evaluation also includes: Perform response time fitting on the timing prediction results based on the collaborative scheduling solution, and establish response speed feedback evaluation based on the response time fitting results; Calculate the frequency recovery rate and frequency deviation control of the coordinated scheduling scheme, and establish response accuracy feedback evaluation based on the calculation results; Conduct stability analysis on the collaborative scheduling scheme and establish response stability feedback evaluation; The response effect evaluation is generated by response speed feedback evaluation, response accuracy feedback evaluation and response stability feedback evaluation.

4. The cross-regional virtual power plant frequency regulation resource coordinated scheduling optimization method combined with a large model as claimed in claim 2 is characterized in that: The cost evaluation also includes: Conduct frequency regulation resource cost analysis of energy types under the coordinated dispatching scheme based on label identification and establish energy cost evaluation; Conduct distance-based cost analysis based on cross-regional connections of collaborative scheduling solutions and establish scheduling cost evaluation; Conduct attenuation cost analysis of frequency modulation resources for the coordinated scheduling scheme and establish attenuation cost evaluation; The cost evaluation is generated by energy cost evaluation, dispatch cost evaluation and attenuation cost evaluation.

5. The cross-regional virtual power plant frequency regulation resource coordinated scheduling optimization method combined with a large model as claimed in claim 2 is characterized in that: The long and short timing adaptation evaluation also includes: Analyze the collaborative scheduling scheme and establish short-term collaborative scheduling scheme and long-term collaborative scheduling scheme; Conduct scheme consistency evaluation on short-term coordinated scheduling schemes and long-term coordinated scheduling schemes, and establish scheme consistency evaluation results; The scheme consistency evaluation results are used to generate the long-short timing adaptation evaluation of the corresponding collaborative scheduling scheme.

6. The cross-regional virtual power plant frequency regulation resource coordinated scheduling optimization method combined with a large model as claimed in claim 1 is characterized in that: The activation adversarial network performs data expansion on the historical data sets of each region, and further includes: After data cleaning of historical data sets, data richness missing analysis is performed to establish data demand space and define data boundaries with historical data sets; The guided noise is established in the data demand space, and the cleaned historical data set is used as the basic data. The data generation network is performed based on the basic data and the guided noise. The discriminant network is initialized through the data boundary, and the generated data is discriminated by the initialized discriminant network; Perform alternating optimization of the generation network and the discrimination network to complete data expansion.

7. The cross-regional virtual power plant frequency regulation resource coordinated scheduling optimization method combined with a large model as claimed in claim 1 is characterized in that: The performing of time series demand prediction of the real-time demand data set and establishing time series prediction results also includes: Configure the window space of short time window and long time window; Use the real-time demand data set to reconstruct short-time windows and long-time windows in the window space and establish the reconstruction results; The reconstruction results are used to predict the demand similarity of the real-time demand data set based on the historical data set after data expansion, and the time series prediction results are established.

8. The cross-regional virtual power plant frequency regulation resource coordinated scheduling optimization method combined with a large model as claimed in claim 1 is characterized in that: The method further comprises: Record the coordinated dispatch data of power plant frequency regulation resources, establish a coordinated dispatch database, and identify the coordinated dispatch database based on dispatch feedback; Use scheduling feedback to build deep collaborative associations, and update the multi-dimensional resource sharing architecture based on deep collaborative associations.

Images